Vortex street characteristic frequency band extraction system based on differential processing of two-path piezoelectric signals

Abstract

The application relates to the technical field of signal processing, and discloses a vortex street characteristic frequency band extraction system based on differential processing of double-path piezoelectric signals, which comprises the following steps: acquiring double-path sampling values, constructing differential mode and common mode signals through charge domain normalization and analytic transformation, and carrying out time-frequency transformation to obtain differential mode and common mode time-frequency spectra; through estimation of leakage coefficients of the common mode to the differential mode time-frequency spectrum, vortex street residual spectrum and modal residual quantity are separated out; frequency band density is constructed by combining the residual spectrum and the modal residual quantity in the time dimension, the frequency band density is mapped into a frequency band value to search for a minimum value and two side boundary points, and a target frequency band is established; finally, a frequency trajectory and a comprehensive bandwidth are calculated by taking the residual spectrum as a weight in the frequency band, and a reconstructed signal is generated; the application effectively suppresses extraction deviation caused by same-direction mechanical crosstalk, and improves the accuracy of signal processing under complex working conditions.

Description

Technical Field

[0001] This invention relates to the field of signal processing technology, and more specifically, to a vortex shedding characteristic frequency band extraction system based on differential processing of dual-channel piezoelectric signals. Background Technology

[0002] In industrial flow measurement applications, vortex shedding detection devices based on piezoelectric sensing commonly employ a dual-piezoelectric element configuration to acquire the alternating signal generated by the alternating shedding of fluid. Under actual operating conditions, the measured pipeline is often in a complex network environment with high-intensity vibration, pump pressure pulsation, or transient valve opening and closing. The detection probe not only senses anti-symmetric pressure pulsations generated by fluid vortices but also simultaneously receives a large amount of co-directional mechanical vibration and fluid impact interference. Engineering applications typically employ hardware or software-level dual-channel differential computation, or a combination of fixed bandpass filtering, to attempt to cancel out this co-directional common-mode disturbance. However, in the actual manufacturing, assembly, and subsequent operation of the sensor and piezoelectric element, the two piezoelectric units inevitably exhibit inherent and dynamically changing microscopic differences in charge conversion sensitivity, cable distributed capacitance, and local strain response gradient. These differences prevent the external co-directional common-mode vibrations from completely canceling each other out after conventional differential computation; instead, they are transformed into differential-mode interference that leaks into the differential channel, forming significant pseudo-frequency band residues. When the flow velocity within the pipe is in a low range or when the flow field pulsates, causing frequency band drift, the interference spectrum caused by residual leakage is easily superimposed on the frequency points of the actual vortex shedding signal. Since conventional signal processing methods mostly rely on fixed amplitude thresholds or monotonically searching for the location of the maximum spectral peak, the system is prone to miscapturing this residual sidelobe caused by differences in strain response as the target frequency band. This often leads to an increase in the dead zone of low-velocity measurements, a sudden increase in measurement deviation under pulsating conditions, or even output stagnation, making it difficult to meet the engineering requirements of stable extraction of weak and non-stationary fluid signals in today's complex industrial environments. Summary of the Invention

[0003] This invention provides a vortex shedding characteristic frequency band extraction system based on differential processing of dual-channel piezoelectric signals, which solves the technical problems mentioned in the background art.

[0004] This invention provides a vortex shedding characteristic frequency band extraction system based on differential processing of dual-channel piezoelectric signals, applied to a flow measurement device comprising a vortex shedding generator, a detection elastic body, and dual-channel piezoelectric elements, configured to perform: The dual-channel sampling sequence output by the dual-channel piezoelectric element is obtained. In order to eliminate the electrical and physical deviation of the channel, the dual-channel sampling sequence is normalized and analytically transformed in the charge domain to obtain the dual-channel analytical signal. Based on the dual-channel analytical signals, differential mode signals and common mode signals are constructed, and time-frequency transformations are performed on them respectively to obtain the differential mode time spectrum carrying antisymmetric fluid pressure and the common mode time spectrum carrying unidirectional mechanical input; Estimate the leakage coefficient of the common-mode time spectrum to the differential-mode time spectrum, and based on the leakage coefficient, separate the vortex shedding residual spectrum stripped of common-mode mechanical interference and the modal residual quantity characterizing structural stress crosstalk; By combining the broadening factor of the residual spectrum of the vortex street in the time dimension used to suppress transient shocks and the modal residual quantity, a frequency band density characterizing the energy distribution of the detached vortex street is constructed. The frequency band density is mapped to the frequency band cost value. By searching for the minimum point of the frequency band cost value, as well as the left and right boundary points that characterize the reverse distribution of frequency band energy, the target frequency band for eliminating structural mode side lobes is established. Within the target frequency band, the frequency trajectory for tracking fluid pulsation characteristics and the comprehensive bandwidth including dispersion and drift information are calculated using the residual spectrum of the vortex shedding as weights. Frequency domain weights are generated based on the frequency band density. Frequency domain weights are then used to perform frequency domain filtering and inverse time-frequency transformation on the vortex shedding residual spectrum to generate a reconstructed signal. The reconstructed signal, the target frequency band, the frequency trajectory, the integrated bandwidth, the modal residual, and the frequency band density are output as a feature set.

[0005] The beneficial effects of this invention are as follows: by normalizing and analyzing the dual signals in the charge domain, the differences in charge conversion sensitivity and wiring between the front-end channels are eliminated, and the leakage coefficient from common mode to differential mode is explicitly constructed in the time-frequency domain, directly eliminating the residual sidelobe interference caused by co-directional inputs such as pipeline mechanical vibration and water hammer; by introducing a thresholdless frequency band delimitation mechanism based on energy density and broadening factor, the system can adaptively track the frequency drift trajectory of fluid pulsation under non-stationary operating conditions, effectively avoiding the phenomenon of misjudging co-directional common-mode mechanical noise as the real vortex street frequency band, and improving the accuracy of signal extraction and the system's anti-crosstalk capability in complex measurement environments in industrial sites. Attached Figure Description

[0006] Figure 1 This is a flowchart of the vortex shedding feature frequency band extraction system based on dual-channel piezoelectric signal differential processing of the present invention. Detailed Implementation

[0007] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0008] like Figure 1As shown, a vortex shedding characteristic frequency band extraction system based on differential processing of dual-channel piezoelectric signals is applied to a flow measurement device that includes a vortex shedding generator, a detection elastic body, and dual-channel piezoelectric elements, and is configured to perform: The dual-channel sampling sequence output by the dual-channel piezoelectric element is obtained. In order to eliminate the electrical and physical deviation of the channel, the dual-channel sampling sequence is normalized and analytically transformed in the charge domain to obtain the dual-channel analytical signal. Based on the dual-channel analytical signals, differential mode signals and common mode signals are constructed, and time-frequency transformations are performed on them respectively to obtain the differential mode time spectrum carrying antisymmetric fluid pressure and the common mode time spectrum carrying unidirectional mechanical input; Estimate the leakage coefficient of the common-mode time spectrum to the differential-mode time spectrum, and based on the leakage coefficient, separate the vortex shedding residual spectrum stripped of common-mode mechanical interference and the modal residual quantity characterizing structural stress crosstalk; By combining the broadening factor of the residual spectrum of the vortex street in the time dimension used to suppress transient shocks and the modal residual quantity, a frequency band density characterizing the energy distribution of the detached vortex street is constructed. The frequency band density is mapped to the frequency band cost value. By searching for the minimum point of the frequency band cost value, as well as the left and right boundary points that characterize the reverse distribution of frequency band energy, the target frequency band for eliminating structural mode side lobes is established. Within the target frequency band, the frequency trajectory for tracking fluid pulsation characteristics and the comprehensive bandwidth including dispersion and drift information are calculated using the residual spectrum of the vortex shedding as weights. Frequency domain weights are generated based on the frequency band density. Frequency domain weights are then used to perform frequency domain filtering and inverse time-frequency transformation on the vortex shedding residual spectrum to generate a reconstructed signal. The reconstructed signal, the target frequency band, the frequency trajectory, the integrated bandwidth, the modal residual, and the frequency band density are output as a feature set.

[0009] The vortex shedding characteristic frequency band extraction system based on dual-channel piezoelectric signal differential processing provided in this embodiment is applied to a flow measurement device that includes a vortex shedding body, a detection elastomer, and dual-channel piezoelectric elements. The flow measurement device uses the vortex shedding body to generate alternating vortices in the fluid. The detection elastomer senses the pressure pulsations generated by the vortices, and the dual-channel piezoelectric elements convert the pressure pulsations into electrical signals, which are then output to the system for processing. The basic signal acquisition parameters used by the system are as follows: the sampling rate fs ranges from 1kHz to 100kHz, determined according to the range of the vortex flowmeter, preferably 10kHz; the sampling bit depth is 16 bits or 24 bits, preferably 24 bits; the processing frame length N ranges from 256 to 4096, preferably 1024; and the frame sliding overlap rate ranges from 0% to 75%, preferably 50%, meaning the number of overlap points between two adjacent frames is N / 2.

[0010] The process involves acquiring the dual-channel sampling sequence output from the dual-channel piezoelectric element, performing normalization and analytical transformation in the charge domain to obtain the dual-channel analytical signal, including the following steps: S201: Based on the physical properties of charge conversion, the dual-path sampling sequence is converted into the first charge quantity and the second charge quantity under the same dimension by calibration coefficient.

[0011] The calculation formula is: ; Where i is the channel index and n is the sampling point index, with values ​​ranging from 0 to N-1; The value of the i-th sequence in the dual-channel sampling sequence is expressed in volts. The calibration coefficient for the i-th channel is obtained by calibration using a laboratory standard flow meter. It characterizes the sensitivity of the i-th channel piezoelectric element in converting charge into voltage, and its value ranges from 1e-9 coulombs / volt to 1e-6 coulombs / volt. Let be the charge of the i-th path, in coulombs.

[0012] Calibration coefficient The calibration was obtained through a combination of static and dynamic calibration in the laboratory. The static calibration steps were as follows: the elastomer was placed in a standard pressure calibration device, and a sinusoidal alternating pressure of known amplitude was applied, with the frequency covering the operating frequency range of the vortex flowmeter; the output voltage sequence of the dual piezoelectric elements was then acquired. The initial calibration coefficient is obtained by calculating the ratio of the root mean square (RMS) value of the output voltage to the RMS value of the applied pressure. The dynamic calibration steps are as follows: Install the assembled vortex flowmeter on a standard flow device, introduce a fluid with a known flow rate, and collect the output voltage sequence of the dual piezoelectric elements; calculate the theoretical vortex frequency and theoretical charge based on the standard flow rate; adjust the initial calibration coefficient so that the frequency of the calculated charge matches the theoretical vortex frequency, and finally determine... The value of the calibration coefficient is valid for 12 months, or it can be recalibrated after the flow meter has been repaired or disassembled.

[0013] S202: To eliminate static baseline drift, extract the first mean of the first charge and the second mean of the second charge within the current processing frame.

[0014] The calculation formula is: ; Where N is the number of sampling points in the current processing frame; is the mean of the i-th path; k is the index of the sampling point, ranging from 0 to N-1.

[0015] S203: By means subtraction and root mean square scaling, the first charge and the second charge are converted into a first normalized value and a second normalized value to eliminate amplitude difference deviation.

[0016] The calculation formula is: ; in, This is the normalized value of the i-th channel, and it is a dimensionless quantity. Anomaly handling mechanism: When the root mean square value in the denominator is less than the preset threshold 1e-6, it indicates that the amplitude of the current frame signal is too low. All normalized values ​​of this channel are set to zero to avoid division by zero errors.

[0017] Preset threshold The value of is determined as follows: when the root mean square value is less than the threshold, the signal-to-noise ratio of the current frame signal is less than 10dB, the effective signal components are submerged by noise and cannot be processed subsequently. Therefore, all normalization values ​​are set to zero to avoid introducing incorrect calculation results.

[0018] S204: Perform a Hilbert transform on the first normalized value to generate a first orthogonal component, and perform a Hilbert transform on the second normalized value to generate a second orthogonal component.

[0019] The Hilbert transform is implemented using the FFT method. The specific steps are as follows: Perform an N-point FFT on a normalized sequence of length N to obtain a frequency domain sequence; set the amplitudes of negative frequency components to zero and multiply the amplitudes of positive frequency components by 2; perform an N-point IFFT on the processed frequency domain sequence to obtain an analytic signal sequence; the imaginary part of the analytic signal sequence is taken as the orthogonal component. Zero-padding is used for boundary processing: N / 2 zeros are padded at the beginning and end of the normalized sequence, and after performing the FFT and IFFT operations, the middle N points are truncated as the final result to reduce the influence of boundary effects.

[0020] For length of normalized value sequence , The frequency domain sequence obtained by point FFT operation is ,in This is a frequency index, with a value range of [value range missing]. to The frequency index corresponding to the negative frequency component. from to The positive frequency component corresponds to the frequency index. from to The part corresponding to the DC component The exact steps for implementing the Hilbert transform are as follows: [The text abruptly ends here, likely due to an incomplete sentence or a formatting error.] conduct Point FFT operation yields ;Will and Keep it unchanged; will Multiply ,in ;Will Set to zero, where ; Perform processing on the frequency domain sequence Pointed IFFT operation yields an analytic signal sequence. ;Pick The imaginary part is the orthogonal component.

[0021] S205: Combine the first normalized value and the first quadrature component to form the first analytical value, and combine the second normalized value and the second quadrature component to form the second analytical value. The first analytical value and the second analytical value together constitute a dual-channel analytical signal.

[0022] The calculation formula is: ; Where j is the imaginary unit, For Hilbert transformation operators; Let be the resolution of the i-th channel, which is a complex sequence. The resolutions of the 1st and 2nd channels together constitute the dual-channel analytic signal. The amplitude of the resolution corresponds to the instantaneous amplitude of the original normalized sequence, and the phase corresponds to the instantaneous phase of the original normalized sequence.

[0023] The differential-mode signal and common-mode signal are constructed based on the dual-channel analytical signal, and time-frequency transformation is performed on them respectively to obtain the differential-mode time spectrum and common-mode time spectrum, including the following steps: S301: Utilizing the reverse cancellation characteristic, the differential mode signal carrying the antisymmetric fluid vortex pressure is extracted from the first and second analytical quantities contained in the dual analytical signals.

[0024] The calculation formula is: ; in, and These are the first and second resolutions, respectively; The signal is a differential-mode signal, a complex sequence. The pressure pulsations generated by fluid vortices on both sides of the detected elastic body are antisymmetric, thus amplified in the differential-mode signal, while mechanical vibrations input in the same direction are suppressed. Anomaly handling mechanism: When and When the amplitude is less than the preset threshold 1e-6, the differential signal is set to zero.

[0025] S302: Utilizing the same-direction superposition characteristic, the common-mode signal carrying the same-direction pipeline mechanical vibration input is extracted from the first and second analytical quantities contained in the dual-path analytical signals.

[0026] The calculation formula is: ; in, The signal is a common-mode signal, a complex sequence. The pressure pulsations generated by the mechanical vibration of the pipeline on both sides of the detected elastic body are distributed in the same direction, thus being amplified in the common-mode signal, while the anti-symmetric fluid vortex pressure is suppressed in the common-mode signal. Anomaly handling mechanism: When and When the amplitude of the signal is less than the preset threshold 1e-6, the common-mode signal is set to zero.

[0027] S303: To balance transient response and frequency resolution, a dynamic window width continuous transform kernel is used to perform time-frequency joint mapping on the differential signal, generating a differential time spectrum containing a complex matrix.

[0028] The dynamic window width continuous transform kernel is essentially a Gaussian window short-time Fourier transform, where the window width is inversely proportional to the frequency. This achieves high frequency resolution in the low-frequency range and good transient response in the high-frequency range. Discretization implementation method: Discrete differential mode signals... Converted to a continuous-time signal via zero-order hold interpolation The sampling interval is Time and location The discretization step size is A total of N time positions are generated; the discretization range of the analysis frequency f is... The frequency step size is A total of M frequency points were generated, of which The continuous integration uses the trapezoidal rule for numerical integration, with an integration step size of . .

[0029] The calculation formula is: ; in, This is the continuous-time equivalent representation of the differential signal; t is the time variable. f represents the time position; f represents the analysis frequency. Let f be the differential mode time spectrum, and let f be a complex matrix of dimension N×M. Anomaly handling mechanism: When f=0, the transform kernel value is zero, and the calculation for that frequency point is skipped.

[0030] The standard deviation of the Gaussian window corresponding to the dynamic window width continuous transformation kernel is: ; This formula is derived from the Gaussian term in the transform kernel, ensuring energy normalization of the Gaussian window. Discrete sequences Convert to continuous time signal The zero-order preserve interpolation formula is: ; in The sampling interval is denoted as .

[0031] For any continuous function In the interval The integral over the given area, using the trapezoidal rule for discrete calculation, is as follows: ; in The sampling interval is... This formula applies to integral operations in all time and frequency dimensions in this embodiment.

[0032] S304: Employs a dynamic window width continuous transform kernel to perform time-frequency joint mapping on the common-mode signal, generating a common-mode time spectrum containing a complex matrix.

[0033] The discretization implementation method is completely consistent with the differential mode time spectrum.

[0034] The calculation formula is: ; in, This is the continuous-time domain equivalent representation of the common-mode signal; Let f be the common-mode frequency spectrum, and let f be a complex matrix of dimension N×M. Anomaly handling mechanism: When f=0, the transform kernel value is zero, and the calculation for that frequency point is skipped.

[0035] Estimate the leakage coefficient of the common-mode spectrum to the differential-mode spectrum, and separate the vortex shedding residual spectrum and modal residuals based on the leakage coefficient, including the following steps: S401: To quantify the modal coupling effect, calculate the cross-power integral of the differential-mode time spectrum and the common-mode time spectrum along the time dimension, as well as the self-power integral of the common-mode time spectrum along the time dimension.

[0036] Discretization implementation method: The trapezoidal numerical integration method is adopted, with an integration step size of . .

[0037] The calculation formula is: ; ; in, The complex conjugate of the common-mode frequency spectrum; The cross-power integral is used to measure the correlation between the differential-mode time spectrum and the common-mode time spectrum at each frequency point. This is the self-power integral, used to measure the energy of the common-mode spectrum at each frequency point. Anomaly handling mechanism: When... When the value is less than the preset threshold 1e-12, it indicates that the common-mode signal energy at that frequency point is too low. Set to zero.

[0038] Preset threshold The value of is determined by: when When the frequency is below this threshold, the common-mode signal energy at that frequency point is lower than the system noise floor, and there is no effective common-mode vibration input. Therefore, Set to zero to avoid numerical instability in subsequent leakage coefficient calculations.

[0039] S402: Based on the mapping ratio of mutual power integral and self-power integral, obtain the leakage coefficient that reflects the proportion of common-mode vibration leakage into the differential channel.

[0040] For scalars, the pseudo-inverse matrix operation is equivalent to the reciprocal operation, specifically implemented as follows: when When greater than or equal to the preset threshold 1e-12, ;when When the value is less than the preset threshold 1e-12, .

[0041] The calculation formula is: ; Where + denotes pseudo-inverse matrix operation; The leakage coefficient is a complex number, whose amplitude represents the proportion of common-mode vibration leaking into the differential channel, ranging from 0 to 1; its phase represents the phase shift of the common-mode vibration in the differential channel, ranging from... to .

[0042] S403: Subtract the product of the leakage coefficient and the common-mode time spectrum from the differential-mode time spectrum to obtain the residual vortex street spectrum stripped of common-mode mechanical interference.

[0043] The calculation formula is: ; in, The residual spectrum of the vortex street is a complex matrix of dimension N×M, mainly containing pure vortex street signal components and a small amount of broadband noise. Anomaly handling mechanism: When When the value is greater than 1, it indicates that the common-mode signal energy is greater than the differential-mode signal energy. At this time, the residual spectrum of the vortex shedding corresponding to that frequency point is set to zero to avoid generating negative energy values.

[0044] when Greater than When the frequency is such that the leakage energy of the common-mode signal in the differential channel is greater than the energy of the original differential-mode signal, the differential-mode signal mainly consists of common-mode leakage and there are no effective vortex shedding signal components. Therefore, the vortex shedding residual spectrum corresponding to this frequency point is set to zero. Threshold The value is based on the theoretical maximum value. ,Exceed This indicates that the calculation result is abnormal.

[0045] S404: Calculate the continuous expansion phase and amplitude of the leakage coefficient, extract the derivative of the continuous expansion phase along the frequency dimension and combine it with the amplitude to obtain the absolute residual quantity reflecting the characteristics of the structural stress gradient change.

[0046] The continuous phase expansion was calculated using the Goldstein phase expansion algorithm, with a tolerance value set to... The expansion direction is from low frequency to high frequency. The specific steps are: first, calculate the principal phase of the leakage coefficient. , range to Between; then starting from the lowest frequency point, compare the phase difference of adjacent frequency points sequentially. If the absolute value of the phase difference is greater than Then add or subtract the phase of the next frequency point. until the absolute value of the phase difference is less than Repeat this process until the phase at all frequency points has been expanded. The phase derivative is calculated using the central difference method with a step size of [value missing]. ,Right now For the first and last frequency points, the forward difference method and the backward difference method are used for calculation respectively.

[0047] The calculation formula is: ; in, The continuous expansion phase of the leakage coefficient; This is the absolute residual value; a larger value indicates a more drastic change in the structural stress gradient at that frequency, and a greater likelihood of structural modal sidelobe interference. Anomaly handling mechanism: When the phase difference between adjacent frequency points is greater than... If the phase expansion fails, the absolute residual at that frequency point is set to zero.

[0048] The continuous phase expansion uses the Goldstein phase expansion algorithm, and the precise steps are: calculating the leakage coefficient. Principal phase , range Initialize continuous expansion phase From the lowest frequency point Begin by processing each pair of adjacent frequency points sequentially. ; Calculate the phase difference ;if Then minus ;if Then Plus Repeat the above steps until all frequency points have been processed. Set the tolerance value to [value to be filled in]. Ensure that the absolute value of the phase difference does not exceed .

[0049] S405: Normalize the absolute residual quantity according to the background level of the analysis frequency band to obtain the modal residual quantity used as the modal sidelobe penalty term for identification.

[0050] Analysis of the lower frequency bound of the frequency band With upper frequency bound Determined based on the range and diameter of the vortex flow meter. The vortex shear frequency corresponding to the lowest flow velocity. This refers to the vortex frequency corresponding to the highest flow velocity. For example, for a DN50 vortex flow meter with a flow range of 1 to 100 cubic meters per hour, the corresponding vortex frequency range is 1 Hz to 1000 Hz. , .

[0051] The calculation formula is: ; in, For frequency integral variables; The modal residual is a dimensionless quantity. A value greater than 1 indicates that the structural residual at that frequency point is higher than the average level, requiring penalty. Anomaly handling mechanism: When the average absolute residual within the analysis frequency band is less than the preset threshold 1e-6, it indicates that the structural modal interference is very small. The modal residuals at all frequency points are set to 1 to avoid the initial density value being too large due to an excessively small denominator.

[0052] By combining the broadening factor of the vortex shedding residual spectrum in the time dimension and the modal residual, the frequency band density is constructed, including the following steps: S501: Accumulate the energy of the vortex shedding residual spectrum along the time dimension to obtain the in-band residual value at each analysis frequency point.

[0053] Discretization implementation method: The trapezoidal numerical integration method is adopted, with an integration step size of . .

[0054] The calculation formula is: ; in, The residual value is the in-band value, representing the total energy of the vortex street residual spectrum at each frequency point.

[0055] S502: To suppress non-periodic transient fluid impact interference, the corresponding broadening factor is calculated based on the time distribution of the vortex shedding spectrum.

[0056] Discretization method: The fourth integral is calculated using the trapezoidal rule for numerical integration, with an integration step size of . ; The time sampling interval is... .

[0057] The calculation formula is: ; in, The broadening factor is a dimensionless quantity used to measure the concentration of the vortex shedding residual spectrum over time. A larger broadening factor indicates a more continuous signal in time, and is more likely to be a periodic vortex shedding signal; a smaller broadening factor indicates a more concentrated signal in time, and is more likely to be transient impulse interference. The broadening factor ranges from 0 to N. Anomaly handling mechanism: When the fourth-power integral in the denominator is less than the preset threshold 1e-24, it indicates that the signal energy at that frequency point is too low, and the broadening factor is set to zero.

[0058] Broadening factor The range of values ​​is to Its physical meaning is the number of sampling points for the equivalent duration of the signal. For a pure sinusoidal signal, which is uniformly distributed in the time dimension, the broadening factor is equal to the processing frame length. For a single impulse signal, which is concentrated at a single sampling point in the time dimension, the broadening factor is equal to... For transient impact signals, the broadening factor is typically smaller than 1. Therefore, transient impact interference can be suppressed by setting a broadening factor threshold.

[0059] S503: Combine the in-band residual with the broadening factor and introduce the modal residual as a denominator penalty term to suppress the residual frequency of the structural gradient and generate the initial density value.

[0060] The calculation formula is: ; in, The initial density value is given. The modal residual is used as a penalty term in the denominator. When the modal residual at a certain frequency is large, the initial density value will be reduced accordingly, thereby suppressing the influence of the structural modal sidelobes.

[0061] S504: The initial density value is calculated according to the distribution ratio of the total energy within the analysis frequency band to obtain the frequency band density that reflects the energy accumulation state of the pure vortex street.

[0062] Discretization implementation method: The trapezoidal numerical integration method is used to calculate the total integral of the initial density value, with an integration step size of . .

[0063] The calculation formula is: ; in, , where is the frequency band density, representing the proportion of pure vortex street signal energy per unit frequency.

[0064] Mapping frequency band density to frequency band cost, and establishing the target frequency band by searching for minimum points and left and right boundary points, includes the following steps: S601: By using a logarithmic inverse mapping, the bandwidth density is transformed into a bandwidth cost that characterizes the divergence trend of energy distribution.

[0065] For frequency step size, .

[0066] The calculation formula is: ; in, Let be the bandwidth cost, a dimensionless quantity. The logarithmic inverse mapping transforms high values ​​of bandwidth density into low values ​​of bandwidth cost, thus allowing the location of the frequency point with the highest vortex shelving energy by finding the minimum point of the bandwidth cost. Anomaly handling mechanism: When When the value is less than the preset threshold 1e-12, the bandwidth cost is set to a large constant, such as 100, to avoid the independent variable of the logarithmic function approaching zero and causing the calculation result to overflow.

[0067] Preset threshold The value of is determined by: when When the energy is below this threshold, the proportion of pure vortex street energy at this frequency point is lower than... It can be ignored.

[0068] constant The value is determined based on: corresponding It is much smaller than the system noise level and can effectively suppress low-energy interference frequencies.

[0069] S602: Within the analysis frequency band, find the minimum point of the frequency band cost value that represents the most dense feature of the vortex street energy distribution.

[0070] The first derivative of discrete data is calculated using the central difference method, with a step size of . The second derivative is also calculated using the central difference method. The criteria for determining the minimum point are: the first derivative is equal to zero and the second derivative is greater than zero.

[0071] The calculation formula is: ; in, The minimum point is defined as follows: Anomaly handling mechanism: When multiple minimum points exist, the minimum point with the lowest corresponding frequency band cost is selected as the final minimum point; when no minimum point satisfies the conditions, the frequency point with the lowest frequency band cost is selected as the minimum point.

[0072] The first derivative of discrete data is calculated using the central difference method, and the formula is: ; For the first and last frequency points, the forward difference method and the backward difference method are used respectively: ; ; The second derivative of discrete data is calculated using the central difference method, and the formula is: ; in This is the frequency point index, with a value range of [value range missing]. to , This represents the frequency step size.

[0073] S603: Locate the physical barrier feature that limits the diffusion of vortex street energy along the direction of decreasing frequency, and use it as the left boundary point.

[0074] The location of the physical barrier characteristic corresponds to the maximum value of the frequency band cost. The criteria for determining the maximum value are: the first derivative is equal to zero and the second derivative is less than zero.

[0075] The calculation formula is: ; in, `sup` represents the left boundary point; `sup` represents the supremum operation. Anomaly handling mechanism: If no maximum point satisfying the conditions is found in the frequency decreasing direction, the lower bound of the analyzed frequency band will be used. As the left boundary point; when there are multiple maxima, select the maxima closest to the minimum point as the left boundary point.

[0076] S604: Locate the physical potential barrier characteristic position that restricts the diffusion of vortex street energy along the direction of increasing frequency, and take it as the right boundary point.

[0077] The calculation formula is: ; in, `inf` represents the right boundary point; `inf` is the infimum operation. Exception handling mechanism: If no maximum point satisfying the conditions is found in the frequency increase direction, the upper bound of the analysis frequency band will be used. As the right boundary point; when there are multiple maximum points, select the maximum point closest to the minimum point as the right boundary point.

[0078] S605: A closed interval is constructed by the left and right boundary points, serving as the target frequency band to eliminate structural mode interference.

[0079] The calculation formula is: ; in, The target frequency band.

[0080] Within the target frequency band, the frequency trajectory and overall bandwidth are calculated using the residual spectrum of the vortex shedding as weights, including the following steps: S701: Within the defined target frequency band, obtain the in-band energy value based on the distribution characteristics of the vortex shelving spectrum.

[0081] Discretization implementation method: The trapezoidal numerical integration method is used to calculate the energy integral within the target frequency band, with an integration step size of . .

[0082] The calculation formula is: ; in, This represents the in-band energy value, indicating the total energy of the residual spectrum of the vortex street within the target frequency band at each time step. Anomaly handling mechanism: When... When the value is less than the preset threshold 1e-12, the frequency trajectory and instantaneous variance at that moment are set to zero.

[0083] S702: Based on the vortex shelving spectrum, the evolution of the transient energy centroid within the target frequency band is calculated, and the frequency trajectory characterizing the fluid pulsation tracking feature is obtained.

[0084] Discretization method: The trapezoidal rule for numerical integration is used to calculate the integrals of the numerator and denominator, with an integration step size of . .

[0085] The calculation formula is: ; in, The frequency trajectory refers to the centroid frequency of energy within the target frequency band at each moment, which can track the frequency changes of the vortex shedding signal in real time.

[0086] S703: Calculate the instantaneous variance that reflects the instantaneous dispersion of the vortex shedding signal based on the degree of dispersion of the energy distribution from the frequency trajectory.

[0087] Discretization method: The trapezoidal rule for numerical integration is used to calculate the integrals of the numerator and denominator, with an integration step size of . .

[0088] The calculation formula is: ; in, This is the instantaneous variance; the larger the value, the more dispersed the frequency distribution of the vortex shear signal at that moment.

[0089] S704: Using the energy value within the band as the weight for time evolution, the time-weighted average of the frequency trajectory is calculated to obtain the center point of the frequency band reflecting the frequency of the main body's shedding.

[0090] Discretization method: The trapezoidal rule for numerical integration is used to calculate the integrals of the numerator and denominator, with an integration step size of . .

[0091] The calculation formula is: ; in, The center point of the frequency band is the average frequency of the vortex shear signal throughout the entire processing frame.

[0092] S705: By combining the instantaneous variance and the drift of the frequency trajectory from the center point of the frequency band, the overall bandwidth characterizing the discrete scale of the main signal is obtained.

[0093] Discretization method: The trapezoidal rule for numerical integration is used to calculate the integrals of the numerator and denominator, with an integration step size of . .

[0094] The calculation formula is: ; in, To achieve comprehensive bandwidth, it also considers the instantaneous dispersion and frequency drift of the signal, thus more accurately reflecting the overall frequency distribution range of the vortex shedding signal.

[0095] Frequency domain weights are generated based on the frequency band density. These weights are then used to perform frequency domain filtering and inverse time-frequency transformation on the vortex shedding residual spectrum to generate a reconstructed signal. The reconstructed signal, target frequency band, frequency trajectory, composite bandwidth, modal residuals, and frequency band density are output as a feature set, including the following steps: S801: Extract the frequency band density within the target frequency band and perform dimensionless transformation according to the proportion within the target frequency band to generate frequency domain weights for extracting pure vortex street components.

[0096] Discretization implementation method: The trapezoidal numerical integration method is used to calculate the total integral of the frequency band density within the target frequency band, with an integration step size of . .

[0097] The calculation formula is: ; in, For frequency domain weights, which are dimensionless quantities; This is an indicator function within the target frequency band, taking a value of 1 within the target frequency band and a value of 0 outside the target frequency band, used to limit the scope of the frequency domain weights. Anomaly handling mechanism: When the integral value in the denominator is less than the preset threshold 1e-12, the frequency domain weights of all frequency points are reset to zero.

[0098] S802: Use frequency domain weights to perform band masking on the residual spectrum of the vortex street, retain the core feature components of the vortex street, and obtain the weighted time spectrum.

[0099] The calculation formula is: ; in, For the weighted time spectrum, it is a complex matrix of dimension N×M, containing only the vortex shear signal components within the target frequency band.

[0100] S803: Inversely transforms the weighted time spectrum back to the time domain to generate a reconstructed signal that eliminates common-mode vibration and broadband noise interference.

[0101] The inverse time-frequency transform employs the inverse Gaussian window short-time Fourier transform corresponding to the forward transform. The discretization implementation method is as follows: for each time position... And frequency point f, weighted time spectrum Multiply by the conjugate dynamic window width continuous transform kernel; integrate along the frequency dimension with an integration step size of . The integral results of all time positions are superimposed to obtain the continuous-time reconstructed signal; the discrete reconstructed signal is obtained by sampling, and its real part is taken as the final result.

[0102] The calculation formula is: ; in, Re is the inverse time-frequency transformation operator, corresponding to the dynamic window width continuous transformation kernel used in steps S303 and S304; Re is the real part sequence operator. The reconstructed signal is a dimensionless real sequence. Boundary processing employs a truncation method, extracting the middle N points of the reconstructed signal as the final result to eliminate the influence of boundary effects.

[0103] The inverse time-frequency transform employs the inverse Gaussian window short-time Fourier transform corresponding to the forward transform, with the specific formula as follows: ; in The total duration of the processed frames, In the discretization implementation, both time integration and frequency integration employ the trapezoidal numerical integration method, with integration step sizes of [missing information]. and .

[0104] S804: Combines the target frequency band, frequency trajectory, integrated bandwidth, modal residual, frequency band density, and reconstructed signal to form a feature set for subsequent traffic calculation and outputs it.

[0105] The calculation formula is: ; in, It is a feature set that includes all of the above items.

[0106] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. A vortex shedding characteristic frequency band extraction system based on differential processing of dual-channel piezoelectric signals, applied to a flow measurement device comprising a vortex shedding generator, a detection elastic body, and dual-channel piezoelectric elements, characterized in that, Configured for execution: The dual-channel sampling sequence output by the dual-channel piezoelectric element is obtained. In order to eliminate the electrical and physical deviation of the channel, the dual-channel sampling sequence is normalized and analytically transformed in the charge domain to obtain the dual-channel analytical signal. Based on the dual-channel analytical signals, differential mode signals and common mode signals are constructed, and time-frequency transformations are performed on them respectively to obtain the differential mode time spectrum carrying antisymmetric fluid pressure and the common mode time spectrum carrying unidirectional mechanical input; Estimate the leakage coefficient of the common-mode time spectrum to the differential-mode time spectrum, and based on the leakage coefficient, separate the vortex shedding residual spectrum stripped of common-mode mechanical interference and the modal residual quantity characterizing structural stress crosstalk; By combining the broadening factor of the residual spectrum of the vortex street in the time dimension used to suppress transient shocks and the modal residual quantity, a frequency band density characterizing the energy distribution of the detached vortex street is constructed. The frequency band density is mapped to the frequency band cost value. By searching for the minimum point of the frequency band cost value, as well as the left and right boundary points that characterize the reverse distribution of frequency band energy, the target frequency band for eliminating structural mode side lobes is established. Within the target frequency band, the frequency trajectory for tracking fluid pulsation characteristics and the comprehensive bandwidth including dispersion and drift information are calculated using the residual spectrum of the vortex shedding as weights. Frequency domain weights are generated based on the frequency band density. Frequency domain weights are then used to perform frequency domain filtering and inverse time-frequency transformation on the vortex shedding residual spectrum to generate a reconstructed signal. The reconstructed signal, the target frequency band, the frequency trajectory, the integrated bandwidth, the modal residual, and the frequency band density are output as a feature set.

2. The vortex shedding characteristic frequency band extraction system based on differential processing of dual-channel piezoelectric signals according to claim 1, which acquires the dual-channel sampling sequence output by the dual-channel piezoelectric sheet and performs normalization processing and analytical transformation in the charge domain to obtain dual-channel analytical signals, includes: Based on the physical properties of charge conversion, the dual-path sampling sequence is converted into a first charge and a second charge under the same dimension by calibration coefficients; To eliminate static baseline drift, the first mean of the first charge and the second mean of the second charge are extracted within the current processing frame, respectively. By subtracting the mean and scaling the root mean square, the first charge and the second charge are converted into a first normalized value and a second normalized value to eliminate the amplitude difference deviation. Perform a Hilbert transform on the first normalized value to generate a first orthogonal component, and perform a Hilbert transform on the second normalized value to generate a second orthogonal component; The first normalized value and the first orthogonal component are combined to form the first analytical quantity, and the second normalized value and the second orthogonal component are combined to form the second analytical quantity. The first resolution and the second resolution together constitute the dual-path resolution signal.

3. The vortex shedding characteristic frequency band extraction system based on dual-channel piezoelectric signal differential processing according to claim 2, comprising constructing differential-mode and common-mode signals based on the dual-channel analytical signals, and performing time-frequency transformations respectively to obtain the differential-mode time spectrum and common-mode time spectrum, including: By utilizing the reverse cancellation characteristic, the differential mode signal carrying the antisymmetric fluid vortex pressure is extracted from the first analytical quantity and the second analytical quantity contained in the dual analytical signal; Utilizing the property of superposition in the same direction, the common-mode signal carrying the mechanical vibration input of the pipeline in the same direction is extracted from the first analytical quantity and the second analytical quantity contained in the dual analytical signal; To balance transient response and frequency resolution, a dynamic window width continuous transform kernel is used to perform time-frequency joint mapping on the differential signal, generating the differential time spectrum containing a complex matrix; The common-mode signal is subjected to a time-frequency joint mapping using the dynamic window width continuous transform kernel to generate the common-mode time spectrum containing a complex matrix.

4. The vortex shedding characteristic frequency band extraction system based on dual-channel piezoelectric signal differential processing according to claim 3, estimating the leakage coefficient of the common-mode time spectrum to the differential-mode time spectrum, and separating the vortex shedding residual spectrum and modal residual quantity based on the leakage coefficient, comprising: To quantify the modal coupling effect, the cross-power integral of the differential mode time spectrum and the common mode time spectrum along the time dimension, as well as the self-power integral of the common mode time spectrum along the time dimension, are calculated. Based on the mapping ratio between the mutual power integral and the self-power integral, the leakage coefficient, which reflects the proportion of common-mode vibration leakage into the differential channel, is obtained. The product of the leakage coefficient and the common-mode time spectrum is subtracted from the differential-mode time spectrum to obtain the vortex shear residual spectrum stripped of common-mode mechanical interference; Calculate the continuous expansion phase and amplitude of the leakage coefficient, extract the derivative of the continuous expansion phase along the frequency dimension and combine it with the amplitude to obtain the absolute residual quantity reflecting the characteristics of structural stress gradient change; The absolute residual is normalized according to the background level of the analysis frequency band to obtain the modal residual used as the modal sidelobe penalty term.

5. The vortex shedding characteristic frequency band extraction system based on dual-channel piezoelectric signal differential processing according to claim 4, comprising constructing a frequency band density by combining the time-dimension broadening factor of the vortex shedding residual spectrum and the modal residual quantity, including: The energy of the vortex shedding residual spectrum is accumulated along the time dimension to obtain the in-band residual value at each analysis frequency point; To suppress non-periodic transient fluid impact interference, the corresponding broadening factor is calculated based on the time distribution of the vortex shedding residual spectrum; The in-band residual value is combined with the broadening factor, and the modal residual is introduced as a denominator penalty term to suppress the residual frequency of the structural gradient, thereby generating the initial density value. The initial density value is calculated according to the distribution ratio of the total energy within the analysis frequency band to obtain the frequency band density that reflects the energy accumulation state of the pure vortex street.

6. The vortex shedding feature frequency band extraction system based on dual-channel piezoelectric signal differential processing according to claim 5, which maps the frequency band density to a frequency band cost and establishes the target frequency band by searching for minimum points and left and right boundary points, includes: By using a logarithmic inverse mapping, the frequency band density is transformed into the frequency band cost value characterizing the deviation trend of energy distribution; Within the analysis frequency band, find the minimum point of the frequency band cost value that represents the most dense feature of the vortex shelving energy distribution; Along the direction of decreasing frequency, locate the physical barrier feature that restricts the diffusion of vortex street energy, and use it as the left boundary point; Along the direction of increasing frequency, locate the physical potential barrier feature that restricts the diffusion of vortex street energy, and use it as the right boundary point; A closed interval is constructed by the left boundary point and the right boundary point, which serves as the target frequency band to eliminate structural mode interference.

7. The vortex shedding feature frequency band extraction system based on dual-channel piezoelectric signal differential processing according to claim 6, wherein within the target frequency band, the frequency trajectory and comprehensive bandwidth are calculated using the residual spectrum of the vortex shedding as weights, comprising: Within the defined target frequency band, the in-band energy value is obtained based on the distribution characteristics of the vortex shedding residual spectrum; Based on the vortex shedding residual spectrum, the evolution of the transient energy centroid within the target frequency band is calculated to obtain the frequency trajectory characterizing the fluid pulsation tracking feature. Based on the degree of dispersion of the energy distribution from the frequency trajectory, the instantaneous variance reflecting the instantaneous dispersion of the vortex shedding signal is calculated; Using the energy value within the band as the time evolution weight, the time-weighted average of the frequency trajectory is calculated to obtain the center point of the frequency band reflecting the frequency of the main body's shedding. By combining the instantaneous variance and the drift of the frequency trajectory from the center point of the frequency band, the overall bandwidth characterizing the overall discrete scale of the main signal is obtained.

8. The vortex shedding feature frequency band extraction system based on dual-channel piezoelectric signal differential processing according to claim 7, comprising generating frequency domain weights based on the frequency band density, performing frequency domain filtering and inverse time-frequency transformation on the vortex shedding residual spectrum using the frequency domain weights to generate a reconstructed signal, and outputting the reconstructed signal, the target frequency band, the frequency trajectory, the comprehensive bandwidth, the modal residual, and the frequency band density as a feature set, including: Extract the frequency band density located within the target frequency band, and perform dimensionless transformation according to the proportion within the target frequency band to generate the frequency domain weights used to extract the pure vortex shedding components; The frequency domain weights are used to perform band masking on the residual spectrum of the vortex shedding to retain the core feature components of the vortex shedding, thus obtaining the weighted time spectrum. The weighted time spectrum is inversely converted back to the time domain to generate the reconstructed signal after removing common-mode vibration and broadband noise interference; The target frequency band, the frequency trajectory, the integrated bandwidth, the modal residual, the frequency band density, and the reconstructed signal are combined to form the feature set for subsequent traffic calculation and then output.

Images