A method for extracting harmonic noise components and a method for evaluating extraction effect

Abstract

The application discloses a harmonic noise component extraction method and an extraction effect evaluation method, which can remove the influence of speed fluctuation by using order tracking, filter high-energy low-frequency interference mixed in the noise by using a filter, solve the problem that the phases of synchronous sections cannot be aligned when synchronous averaging is applied, further improve the signal-to-noise ratio by combining the autocorrelation noise reduction advantage, and evaluate the extraction effect of the harmonic noise component extraction method by using the proposed extraction effect evaluation method, thereby providing a basis for improving the extraction effect. The extraction method comprises the following steps: in an airborne and ground system, noise signals and speed signals are collected under the same working condition; rotation frequency data are obtained by processing; order tracking is performed on the filtered signals by using the rotation frequency data, and the rotation frequency data are converted into angular domain stationary signals; synchronous averaging is performed on the angular domain stationary signals to obtain synchronous average signals; autocorrelation operation and Fourier transformation are performed on the synchronous average signals to obtain order spectrum, and finally the extraction of the harmonic noise component is completed.

Description

Technical Field

[0001] This invention relates to the field of airborne and ground system equipment condition monitoring technology, specifically to a method for extracting harmonic noise components and a method for evaluating the extraction effect. Background Technology

[0002] In airborne systems, vibration sensors are used to monitor the vibration of various aircraft components to detect potential mechanical failures. To meet the requirements of high-temperature environments, high-temperature resistant vibration sensors can be used in conjunction with sound sensors to obtain more comprehensive data. Sound sensors can quickly acquire sound signals (noise signals) in a non-contact manner, providing supplementary information to the vibration data.

[0003] Signal Acquisition and Processing: Handling irrelevant interference and noise components is crucial during signal acquisition. Since mechanical faults primarily manifest as harmonic vibrations, relevant information processing algorithms can be used to extract harmonic components. Time-domain synchronous averaging (TSA) is an effective technique for improving the signal-to-noise ratio, especially when rotational speed information is available. If rotational speed information is unavailable, fractional-order time-domain synchronous averaging (FTSA) can be used, with phase compensation based on cross-power spectrum to address the problem of accumulated phase errors.

[0004] Noise reduction techniques: To further improve signal quality, autocorrelation processing can be considered as a noise reduction technique. This method does not require prior knowledge and can effectively improve the signal-to-noise ratio. In airborne systems, this can help reduce the impact of interference sources such as vibrations from engine-independent components and complex environments, thereby improving the accuracy of fault detection.

[0005] In summary, by combining vibration sensors, sound sensors, and the aforementioned signal processing technologies, airborne and ground systems can more reliably monitor and diagnose mechanical faults, improving flight safety and equipment reliability. The application of these technologies contributes to enhancing the performance and efficiency of aviation systems.

[0006] While time-domain synchronous averaging based on cross-power spectrum effectively solves the phase error problem, it still requires a near-constant rotational frequency. The greater the speed fluctuation, the more drastically the better the method performs. Time-domain synchronous averaging is widely used in vibration signal processing, but rarely applied in sound signal processing. Practical results show that, due to specific differences and limitations, it cannot be directly transferred to noise signal processing. Time-domain synchronous averaging can extract harmonic component information, but its improvement in signal-to-noise ratio is limited.

[0007] Therefore, existing methods for extracting harmonic noise components in airborne and ground systems cannot solve the problems of the influence of speed fluctuations, limited improvement in signal-to-noise ratio, and poor extraction effect. Summary of the Invention

[0008] In view of this, the present invention provides a method for extracting harmonic noise components and an evaluation method for the extraction effect. It can use order tracking to remove the influence of speed fluctuations; apply a filter to remove high-energy low-frequency interference mixed in the noise, and solve the problem of phase misalignment in the synchronization segment when applying synchronous averaging; finally, combine the advantages of autocorrelation noise reduction to further improve the signal-to-noise ratio. At the same time, the proposed extraction effect evaluation method is used to evaluate the extraction effect of the harmonic noise component extraction method, providing a basis for improving the extraction effect.

[0009] To achieve the above objectives, the present invention provides a method for extracting harmonic noise components, comprising the following steps:

[0010] Step 1: Under the same operating conditions, noise signals and rotational speed signals are collected separately in both airborne and ground systems.

[0011] The time interval is set to be at least the time it takes for the rotor to be analyzed to rotate 2 revolutions. The noise signal and speed signal are extracted using the set time interval to obtain a series of short-time signals of equal length and corresponding short-time speed signals.

[0012] Step 2: Process a series of short-time signals of equal duration and corresponding short-time rotational speed signals to obtain frequency conversion data. Then, use the frequency conversion data to perform order tracking on the filtered signal and convert it into a stationary signal in the angular domain.

[0013] Step 3: Perform synchronous averaging on the stationary signal in the angular domain to obtain the synchronous average signal.

[0014] Step 4: The synchronous average signal is subjected to autocorrelation operation to obtain the autocorrelation signal, and then Fourier transform is performed to obtain the order spectrum, thus completing the extraction of harmonic noise components.

[0015] Furthermore, in step one, under the same operating conditions in both airborne and ground systems, noise signals are collected using sound sensors, and rotational speed signals are collected using rotational speed sensors.

[0016] Furthermore, a series of short-time signals of equal duration and corresponding short-time rotational speed signals are processed to obtain frequency data. The specific steps are as follows:

[0017] The series of short-time signals of equal length obtained in step one is x i (t), corresponding to the short-time rotational speed signal n i (t), where i is the sequence number.

[0018] For each short-time speed signal n i (t) Find its average value mn i .

[0019] Short-time signal x i (t) is transformed by Fourier FFT to obtain the spectrum x i (f)

[0020] x i (f) = FFT(x) i (t))

[0021] Where FFT(·) is the Fourier transform, f is the frequency, and the unit is Hz.

[0022] Based on mn i and spectrum x i (f) Obtain frequency conversion data F(i)

[0023] F(i) = nearmax(x) i (f=mn i / 60))

[0024] nearmax(·) is used to calculate the spectrum x. i (f) Distance f = mn i The operation corresponding to the nearest maximum value of / 60 is the frequency value F(i) on the horizontal axis, where F(i) is in Hz.

[0025] First, observe the spectrum x. i (f), according to the spectrum x i (f) For the range of component concentration, set the upper limit frequency fh of the bandpass filter; secondly, set the lower limit frequency fl of the bandpass filter according to the sound field environment, fl is set empirically, and the filtering parameters [fl, fh] are applied to the short-time signal x. i (t) Perform bandpass filtering to obtain the filtered signal xf i (t).

[0026] Define a rectangular window function w(f):

[0027]

[0028] Based on the given filter bandwidth, the spectrum x i (f) is multiplied by the rectangular window function w(f), and then the inverse Fourier transform is performed to obtain the corresponding filtered signal xf. i (t):

[0029] xf i (f)=x i (f)·w(f)

[0030]

[0031] Further, step two: Process a series of short-time signals of equal duration and their corresponding short-time rotational speed signals to obtain frequency conversion data. Then, use the frequency conversion data to perform order tracking on the filtered signal, converting it into a stationary signal in the angular domain. Specifically:

[0032] Based on the filtered signal xf i (t) corresponds to the conversion frequency F(i), and the conversion frequency F(i)×60 is copied and extended to the signal xf respectively. i (t) of the same length as in formula (7), then for xf i (t) Perform order tracking to obtain the corresponding angular domain signal ys i (r), where r is the number of revolutions.

[0033] rpm i =(F(i)×60…F(i)×60…F(i)×60) 1×N

[0034] Where N is the signal xf i The data length of (t).

[0035] Order tracking involves equal-angle sampling, which essentially means acquiring equal-angle moments. After acquiring these moments, interpolation and fitting on the original function yields the stationary signal in the angular domain, i.e., based on the shortest data length angular domain signal ys. i (r) The maximum integer revolutions L contained in all angular domain signals are uniformly truncated, and the truncated equal-length signal is defined as reys. i (n), where n represents the data points of the signal n = 1, 2, 3, ..., reys i (n) is the obtained angular domain stationary signal.

[0036] Further, step three: Perform synchronous averaging on the stationary signal in the angular domain to obtain the synchronous average signal, specifically including the following steps:

[0037] a. For angularly stationary signals reys i (n), the phase detection signal of its i-th segment is defined as:

[0038]

[0039] Where Fs is the resampling frequency set by the order tracking, and its value is equal to the number of sampling points when the reference axis rotates one revolution.

[0040] b. Construct a reference cosine sequence of the same length as the phase detection signal.

[0041]

[0042] Then calculate the cross power spectrum P. i(k) and phase Phase is used for phase compensation:

[0043]

[0044]

[0045] c. Convert the phase obtained in b into the number of sampling points. The expression is:

[0046]

[0047] The round(·) function performs rounding to the nearest integer.

[0048] The phase difference between segments is eliminated by cyclic shifting. This cyclic shifting process can be represented as follows:

[0049]

[0050] in It is the i-th synchronization segment.

[0051] d. Take the average value of all synchronization segments:

[0052]

[0053] Where xc is the final average synchronization signal, and Q is the number of synchronization segments.

[0054] Another embodiment of the present invention provides a method for evaluating the extraction effect of a harmonic noise component extraction method. The harmonic noise component is extracted using the above-mentioned harmonic noise component extraction method. The following three Fourier spectral domain quantitative indicators are used to measure the effectiveness of the harmonic noise component extraction: harmonic significance FFC, proportion of prominent spectral lines FFCnum, and comprehensive index CL.

[0055] Harmonic Significance (FFC):

[0056] First, define the integer prominence multiplier Af. i Judgment criteria: Spectral lines in the order spectrum whose amplitude is greater than the threshold thr and whose corresponding order is an integer are defined as harmonic prominent spectral lines, where the spectral lines are harmonic prominent spectral lines in the spectrum.

[0057]

[0058] Where Af is the full spectrum amplitude, max(·) is the maximum value operation, and mean(·) is the average value operation. i To emphasize the amplitude of the harmonic order spectral lines, Af i The corresponding order is f i N * Represents a positive integer.

[0059] Secondly, the saliency range of the prominent harmonic spectral lines is defined as the open set (f). i -2,f i +2).

[0060] Finally, the harmonic significance index value FFC is calculated according to the formula:

[0061]

[0062] Where f iw For the significance range, Af iw f iw For the corresponding spectral amplitude, sum(·) is the summation operation.

[0063] The proportion of harmonic frequencies of prominent spectral lines FFCnum: In the order spectrum, spectral lines whose amplitude exceeds the threshold thr defined in A are considered prominent spectral lines. When counting the number of prominent spectral lines, if the order values ​​of adjacent prominent spectral lines do not exceed 0.1, they are counted only once.

[0064]

[0065] Where FFCnum is the proportion of harmonic frequencies of prominent spectral lines, Af k To effectively highlight the spectral line amplitude, Af k The corresponding order is f k num(·) is a statistical operation.

[0066] The comprehensive index CL is the sum of the two indices mentioned above:

[0067] CL = FFC + FFCnum

[0068] The comprehensive index CL is the sum of the two indices mentioned above, aiming to provide a balanced representation of the overall performance of the studied method in terms of both the proportion of harmonic frequencies of prominent spectral lines and the significance of harmonic frequencies. This comprehensive index allows for a holistic evaluation of the method's effectiveness and feasibility.

[0069] Beneficial effects:

[0070] 1. This invention provides a method for extracting harmonic noise components, applicable to airborne and ground systems. The invention uses order tracking to remove the influence of speed fluctuations; applies filters to remove high-energy low-frequency interference mixed in with the noise, solving the problem of phase misalignment during synchronization averaging; and finally, combines the advantages of autocorrelation denoising to further improve the signal-to-noise ratio. Furthermore, the proposed harmonic noise component extraction method is capable of handling emergency situations such as discontinuous data acquisition and partial data loss, and the proposed method has extremely fast computation speed. This method can be applied to airborne and ground systems to improve the accuracy and efficiency of data processing, especially when dealing with noise interference and discontinuous data.

[0071] 2. This invention also provides an evaluation method for harmonic noise component extraction, proposing three Fourier spectral domain quantitative indicators to measure the effectiveness of the proposed method. These three Fourier spectral domain quantitative indicators are: harmonic significance, proportion of prominent spectral line harmonics, and a comprehensive index. These three Fourier spectral domain quantitative indicators allow for a more comprehensive evaluation of the method's effectiveness and feasibility. Attached Figure Description

[0072] Figure 1 The flowchart is a method for extracting harmonic noise components provided by the present invention.

[0073] Figure 2 This is an example spectrum diagram from an embodiment of the present invention;

[0074] Figure 3 This is an example diagram of the spectrum of a single group of data under high-speed operating conditions in an embodiment of the present invention;

[0075] Figure 4 This is a comparison of the combustion reference order spectrum in the combustion speed reference example of the present invention; Figure 4 (a) in the figure represents the order spectrum of a single set of data before processing; Figure 4 (b) in the figure represents the order spectrum of the synchronous average signal; Figure 4 (c) in the figure represents the order spectrum of the introduced autocorrelation signal;

[0076] Figure 5 This is a comparison chart of index measurement in the benchmark example of the combustion speed of this invention; the comparison of order spectrum indexes before and after synchronous averaging processing is as follows: Figure 5 As shown in (a) above, the order spectral index comparison before and after introducing autocorrelation treatment is as follows: Figure 5 As shown in (b)

[0077] Figure 6 This is a comparison diagram of the order spectrum of the moving vortex reference in the example of the moving vortex speed reference of the present invention; Figure 6 (a) in the figure represents the order spectrum of the single-rent data before processing; Figure 6 (b) in the diagram represents the order spectrum of the synchronous average signal. Figure 6 In; Figure 6 (c) in the figure represents the order spectrum of the introduced autocorrelation signal;

[0078] Figure 7 This is a comparison chart of index measurement in the reference example of the vortex speed of this invention; a comparison of order spectrum indexes before and after synchronous averaging processing is shown below. Figure 7 As shown in (a) above, the order spectral index comparison before and after introducing autocorrelation treatment is as follows: Figure 7 As shown in (b);

[0079] Figure 8 A comparison diagram of the reference order spectrum of the dynamic vortex in an example where the method proposed in this invention has not been filtered; Figure 8 (a) in the figure represents the order spectrum of the synchronous average signal; Figure 8 (b) in the figure represents the order spectrum of the signal after autocorrelation processing.

[0080] Figure 9 This is a comparison chart of performance metrics in instances where the method proposed in this invention was not filtered. Figure 9 (a) in the figure is a comparison of the indicators before and after synchronous averaging; Figure 9 (b) in the figure is a comparison chart of the indicators before and after the introduction of autocorrelation treatment;

[0081] Figure 10 Comparison of the baseline order spectrum of dynamic eddy currents in the example of time-domain synchronous average (TSA) + autocorrelation; Figure 10 (a) in the figure represents the order spectrum of the synchronous average signal; Figure 10 (b) in the figure represents the order spectrum of the signal after autocorrelation processing.

[0082] Figure 11 A comparison chart of metrics in instances of Time-Domain Synchronous Average (TSA) + Autocorrelation; Figure 11 (a) in the figure is a comparison of the indicators before and after synchronous averaging; Figure 11 (b) in the figure is a comparison chart of the indicators before and after the introduction of autocorrelation treatment. Detailed Implementation

[0083] The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0084] Example 1:

[0085] This invention provides a method for extracting harmonic noise components. The flowchart of this method is shown below. Figure 1 As shown, the specific steps are as follows:

[0086] Step 1: Signal Preprocessing

[0087] In both airborne and ground-based systems, noise and rotational speed signals are collected under the same operating conditions. The time interval is set to be at least two revolutions of the rotor to be analyzed. The noise and rotational speed signals are then extracted using this set time interval to obtain a series of short-time signals of equal length and corresponding short-time rotational speed signals.

[0088] In this embodiment of the invention, signal acquisition is crucial in airborne and ground systems, including the acquisition of noise and rotational speed signals. The noise signal x(t) is acquired using a sound sensor, and the rotational speed signal n(t) is acquired using a rotational speed sensor, where t represents time. Data is extracted at arbitrary time intervals of 0.1 seconds under the same operating conditions (the extraction time should be at least the time it takes for the rotor to be analyzed to rotate twice; the extraction time should not be too long to avoid large fluctuations in rotational speed within the time period). These arbitrary time interval extractions are used to simulate the condition of discontinuous data to be processed, thereby demonstrating the ability of the proposed technology to handle emergency situations involving discontinuous acquisition and partial data loss. Therefore, a series of short-time signals x of equal duration can be obtained. i (t) and the corresponding short-time speed signal n i (t). This signal preprocessing method can be applied in airborne and ground systems, helping to ensure the reliability of the information contained in the signal and providing a solid foundation for subsequent analysis and processing. This is of great significance in improving system performance and security.

[0089] Each short-time speed signal n i (t) Find its average value mn i The reference speed is used to find the actual speed frequency later. The short-term signal speed fluctuation is very small and can be ignored.

[0090] mn i =mean(n) i (t)) (1)

[0091] Where t is time, in seconds; mean(·) is the average value operation; and the rotational speed signal n... i (t) is in r / min.

[0092] Short-time signal x i (t) is transformed by Fourier Transform (FFT) to obtain the spectrum x i (f)

[0093] x i (f) = FFT(x) i (t)) (2)

[0094] FFT(·) is the Fourier transform, where f is the frequency in Hz.

[0095] Based on mn i and spectrum x i (f) Obtain the frequency conversion F(i)

[0096] F(i) = nearmax(x) i (f=mn i / 60)) (3)

[0097] nearmax(·) is used to calculate the spectrum x. i (f) Distance f = mn i The operation corresponding to the nearest maximum value of / 60 is the frequency value F(i) on the horizontal axis, where F(i) is in Hz.

[0098] Figure 2 An example of the spectrum is shown; first observe... Figure 2 The spectrum x in i (f) The main components of the spectrum are concentrated in the range of [0, 25000] Hz, therefore the upper limit frequency of the bandpass filter is set to fh = 25000 Hz; secondly, the lower limit frequency of the bandpass filter is set according to the sound field environment (based on experience, usually below 60 Hz), and the filtering parameters [fl, fh] are applied to the short-time signal x. i (t) Perform bandpass filtering to obtain the filtered signal xf i (t).

[0099] Define a rectangular window function w(f):

[0100]

[0101] Based on the given filter bandwidth, the spectrum x i (f) is multiplied by the rectangular window function w(f), and then the inverse Fourier transform is performed to obtain the corresponding filtered signal xf. i (t).

[0102] xf i (f)=x i (f)·w(f) (5)

[0103]

[0104] Step 2 Order Tracking

[0105] In airborne and ground systems, frequency conversion data is preprocessed using filtered signals, and then the filtered signals are tracked by order using the frequency conversion data to convert them into angular domain stationary signals.

[0106] Based on the filtered signal xf i (t) corresponds to the conversion frequency F(i), and the conversion frequency F(i)×60 is copied and extended to the signal xf respectively. i (t) The same length is shown in Formula 7, and then xf i (t) Perform order tracking to obtain the corresponding angular domain signal ys i(r), where r is the number of revolutions (e.g., F(1)×60 is copied and extended for xf1(t) order tracking, F(2)×60 is copied and extended for xf2(t) order tracking, and so on). The key to order tracking is equal-angle sampling, which is essentially the acquisition of equal-angle moments. After acquiring equal-angle moments, the angular domain stationary signal can be obtained by interpolation and fitting on the original function. [6] .

[0107] This method is very effective in signal processing, and is particularly suitable for scenarios in airborne and ground systems where precise speed tracking and analysis of signals are required.

[0108] rpm i =(F(i)×60…F(i)×60…F(i)×60) 1×N (7)

[0109] Where N is the signal xf i The data length of (t).

[0110] The order tracking method is an existing method. It is based on the shortest data length angular domain signal ys. i (r) The maximum integer revolutions L contained in all angular domain signals are uniformly truncated, and the truncated equal-length signal is defined as reys. i (n), where n represents the data points of the signal n = 1, 2, 3...

[0111] Step 3: Synchronous Averaging

[0112] Synchronous averaging is performed on the stationary signal in the angular domain to obtain the synchronous average signal.

[0113] For continuous signals P1(t) and P2(t), their cross-correlation function is expressed as:

[0114]

[0115] Where τ is the delay time.

[0116] This function is primarily used to reflect the correlation between two different signals. After cross-correlation, components with the same frequency are enhanced, preserving phase information, while other components are suppressed.

[0117] In practical applications, phase detection methods based on cross-power spectra can be employed to better understand and utilize this correlation information. This method allows the present invention to more accurately determine the phase differences between signals and further utilize this information for signal processing and analysis. This is of great value for data processing and decision-making in airborne and ground systems, as it helps improve the understanding and utilization of signals, thereby improving system performance and efficiency.

[0118] To detect the phase of the fundamental frequency component in the discrete signal x(n), a reference cosine sequence is constructed as follows:

[0119] y(n)=cos(2πf0n) (9)

[0120] Where f0 is the fundamental frequency in x(n), in Hz.

[0121] According to the definition of convolution, the cross-correlation function can be expressed as:

[0122] R(n)=x(n)*y(-n) (10)

[0123] Where: * represents convolution.

[0124] Using Fourier transform and the time-domain convolution theorem, the cross-power spectrum can be calculated as follows:

[0125] P(k)=X(k)Y * (k) (11)

[0126] Where Y * (k) is the conjugate complex form of Y(k), and X(k) and Y(k) represent the discrete Fourier transforms of x(n) and y(n).

[0127] Therefore, the fundamental frequency component of the signal is enhanced. Its phase... It equals the phase of a complex number at the maximum amplitude of the cross power spectrum. The expression is:

[0128]

[0129] This method enhances the fundamental frequency component through cross-correlation operations and extracts the phase from the cross-power spectrum. It has advantages such as strong adaptability, high accuracy, and low computational cost.

[0130] The synchronous averaging process in this embodiment of the invention is as follows:

[0131] a. Therefore, the phase detection signal of the i-th segment can be defined as:

[0132]

[0133] Where Fs is the resampling frequency set by the order tracking, and its value is equal to the number of sampling points when the reference axis (the axis to be analyzed) rotates one revolution. The i-th segment of the phase detection signal, where n is discrete data, representing the n-th element in the phase detection signal.

[0134] b. Construct a reference cosine sequence of the same length as the phase detection signal.

[0135]

[0136] Then calculate the cross power spectrum P. i (k) and phase Phase is used for phase compensation.

[0137]

[0138]

[0139] c. Convert the phase obtained in b into the number of sampling points. The expression is:

[0140]

[0141] The round(·) function performs rounding to the nearest integer.

[0142] The phase difference between segments is eliminated by cyclic shifting. This process can be represented as:

[0143]

[0144] in It is the i-th synchronization segment.

[0145] d. Take the average value of all synchronization segments:

[0146]

[0147] Where xc is the final average synchronization signal, and Q is the number of synchronization segments.

[0148] Step 4 Autocorrelation

[0149] The autocorrelation function can retain periodic signals in the signal to be processed and remove non-periodic Gaussian white noise.

[0150] The application of autocorrelation functions helps extract and enhance periodic signals in airborne and ground systems, while effectively reducing the impact of aperiodic noise. This is of great significance for signal analysis, fault detection, and data cleaning, contributing to improved system performance and data quality.

[0151] The synchronous average signal xc is autocorrelation-operated to obtain signal xACF, which is then Fourier transformed to obtain the order spectrum, and finally the harmonic noise components are extracted.

[0152] Example 2:

[0153] Another embodiment of this invention proposes three quantitative indices in the Fourier spectral domain to measure the effectiveness of the proposed method. The following definition uses the order spectrum in the Fourier spectral domain as an example:

[0154] Harmonic Significance FFC

[0155] First, define the integer prominence multiplier Af. i Judgment criteria. Spectral lines in the order spectrum whose amplitude is greater than the threshold thr and whose corresponding order is an integer are defined as harmonic prominent spectral lines (harmonic prominent spectral lines in the spectrum).

[0156]

[0157] Where Af is the full spectrum amplitude, max(·) is the maximum value operation, and mean(·) is the average value operation. i To emphasize the amplitude of the harmonic order spectral lines, Af i The corresponding order is f i N * Represents a positive integer.

[0158] Secondly, the saliency range of the prominent harmonic spectral lines is defined as the open set (f). i -2,f i +2). The significance range is set to 2 on both sides to take into account that the minimum distance between the frequencies of typical turboshaft engine components and the order spectrum in subsequent processing is 2.

[0159] Finally, the harmonic significance index value FFC is calculated based on the formula.

[0160]

[0161] Where f iw For the significance range, Af iw f iw For the corresponding spectral amplitude, sum(·) is the summation operation.

[0162] Highlight the proportion of harmonic frequencies of spectral lines (FFCnum)

[0163] In the order spectrum, spectral lines whose amplitude exceeds the threshold thr defined in A are considered prominent spectral lines. When counting the number of prominent spectral lines, if the order value corresponding to adjacent prominent spectral lines does not exceed 0.1, then only one count is made.

[0164]

[0165] Where FFCnum is the proportion of harmonic frequencies of prominent spectral lines, Af k To effectively highlight the spectral line amplitude, Af k The corresponding order is f k num(·) is a statistical operation.

[0166] Comprehensive index CL

[0167] CL = FFC + FFCnum (23)

[0168] The comprehensive index CL is the sum of the two indices mentioned above, aiming to provide a balanced representation of the overall performance of the studied method in terms of both the proportion of harmonic frequencies of prominent spectral lines and the significance of harmonic frequencies. Through this comprehensive index, the effectiveness and feasibility of the method can be evaluated more comprehensively.

[0169] Verification by measured data

[0170] This invention details the harmonic component extraction process of noise acquisition data during the test run of a certain type of turboshaft engine. This solution can be applied to airborne and ground systems. During a single test run, data is acquired throughout the entire process at a sampling frequency of 102400Hz.

[0171] Specifically, the turboshaft engine adopts a free turbine structure, consisting of a gas generator (gas generator) rotor and a power turbine (power turbine) rotor. In this context, it is necessary to extract the harmonic components of the gas generator rotor and the power turbine rotor from the collected data.

[0172] Data was captured at arbitrary time intervals of 0.1s under high-speed conditions, resulting in 27 sets of data. One set of signal spectrum was selected as follows: Figure 3 As shown.

[0173] Engine speed reference

[0174] The rotational frequency is obtained based on the combustion speed, and then the order tracking signal reys before synchronous averaging is obtained according to the proposed method. i (r) The synchronous average signal xc and the autocorrelation-processed signal xACF are respectively subjected to FFT transformation to obtain the order spectrum as follows: Figure 4 (a) Figure 4 (b) and Figure 4 As shown in (c), the comparison of order spectral indices before and after synchronous averaging is as follows: Figure 5 As shown in (a), the comparison of order spectral indices before and after introducing autocorrelation processing is as follows: Figure 5 As shown in (b).

[0175] Figure 4 (a) The order spectral components are numerous and chaotic, making it difficult to capture key information. Figure 4 (b) The harmonic order components in the order spectrum are significantly enhanced, while irrelevant interference and background noise components are significantly suppressed. Figure 5 (a) The significance of harmonic frequencies was significantly improved; only one of the 27 data sets had an index higher than the signal after synchronous averaging. The proportion of prominent spectral harmonic frequencies was approximately twice that of all unprocessed data. The overall index was approximately twice that of all unprocessed data. This indicates that the synchronous averaging method achieved excellent results.

[0176] Figure 4(c) The order spectrum clearly shows specific order components, and the prominent spectral lines are all harmonic order spectral lines. The two arrows indicate the order components corresponding to the passing frequencies of the two-stage combustion blades. Irrelevant interference and background noise components are significantly and further suppressed. Figure 5 (b) compared to Figure 5 (a) The harmonic significance index has been significantly improved; the proportion of prominent spectral harmonics has not changed significantly; the overall index has finally increased to about 3 times.

[0177] In summary, the proposed method is significantly effective for extracting harmonic noise components.

[0178] Dynamic vortex speed reference

[0179] The rotational frequency is obtained based on the vortex rotational speed, and then the order tracking signal reys before synchronous averaging is obtained according to the proposed method. i (r) The order spectra are obtained by performing FFT transformations on the synchronous average signal xc and the introduced autocorrelation signal xACF, respectively. Figure 6 (a) Figure 6 (b) and Figure 6 As shown in (c), the comparison of order spectral indices before and after synchronous averaging is as follows: Figure 7 As shown in (a), the comparison of order spectral indices before and after introducing autocorrelation processing is as follows: Figure 7 As shown in (b).

[0180] Figure 6 (a) The order spectral components are numerous and chaotic, making it difficult to capture key information. After synchronous averaging, as shown... Figure 6 As shown in (b), the harmonic components related to the vortex speed reference dominate the order spectrum, but there are still irrelevant background noise components. Figure 6 (a) The improvement in harmonic frequency significance is not very obvious; through analysis of the proportion of harmonic frequencies of prominent spectral lines and comprehensive indicators, it was found that only 3 sets of data had higher indicator values ​​than the results of the synchronous average signal, while the proportion of harmonic frequencies of prominent spectral lines was significantly improved in the other 24 sets of data after synchronous averaging. This indicates that the synchronous averaging method has achieved a significant improvement effect.

[0181] Figure 6 (c) The order spectrum clearly shows specific order components. The two arrows in the figure indicate the order components corresponding to the passing frequencies of the two-stage blades of the moving vortex. In addition, the component at harmonic order 15 is also significantly enhanced, while the non-harmonic orders 1.65 and 28.21 and the noise floor component are significantly suppressed. Figure 7 (b) After introducing autocorrelation processing, all three indicator values ​​were higher than the unprocessed data, with the overall indicator improving by more than two times compared to before processing. Figure 7 (a) All three indicators have been further improved.

[0182] The above experimental data verify that the method proposed in this invention has significant effectiveness in extracting harmonic noise components.

[0183] The application of this technology is crucial for monitoring and analyzing the performance and condition of turboshaft engines, whether in airborne or ground-based systems. It helps identify potential problems and faults, improving the reliability and safety of the mechanical system. Furthermore, by extracting harmonic components, a better understanding of the turboshaft engine's operation can be achieved, leading to more precise maintenance and optimization.

[0184] To verify the effectiveness of the proposed method, this invention compared its application with two other methods. The same dataset and evaluation criteria were used to compare the performance of the proposed method with the comparative methods. The comparison results objectively evaluate the performance of different methods in information extraction and determine the advantages of the proposed method in terms of accuracy, robustness, etc.

[0185] To simplify the discussion and facilitate comparative analysis, only the vortex rotation speed reference is used for comparison of the methods proposed in this invention.

[0186] The method proposed in this invention does not perform filtering. The original noise signal is followed by order tracking based on the frequency shift according to the flowchart of the method proposed in this invention. Then, synchronous averaging and autocorrelation processing are performed sequentially. No filtering is performed before synchronous averaging.

[0187] Figure 8 (a) The spectral lines in the order spectrum of the synchronous average signal are densely distributed, showing a chaotic component phenomenon, and the harmonic order components are no longer clearly distinguishable. Figure 8 In (b), the anharmonic order 28.24 component is the most prominent. Figure 9 (a) The harmonic significance index is lower than that of all individual signals before processing. After autocorrelation to improve the harmonic significance, such as Figure 9 As shown in (b), the overall index is still lower than the signal before the first group of processing.

[0188] Based on experiments and data analysis, filtering significantly impacts the effectiveness of synchronous averaging. A crucial prerequisite for synchronous averaging is achieving phase alignment to maximize the superposition of useful signal information during the averaging process. However, for noisy signals, high-energy low-frequency interference components severely affect phase alignment during the synchronization segment.

[0189] Therefore, this invention concludes that filtering out low-frequency interference is crucial for improving the performance and accuracy of synchronization averaging. By filtering out low-frequency interference, unnecessary components in the noise signal can be effectively reduced, thereby mitigating interference with phase alignment during synchronization and thus improving the effectiveness of synchronization averaging. This conclusion is of great significance for improving signal processing and data analysis in both airborne and ground systems, contributing to improved performance and accuracy, especially when dealing with noise interference.

[0190] Time-domain synchronous averaging (TSA) + autocorrelation: Single noise signals are sorted according to their time occurrence order. The frequency shift is taken as the average of the frequency shifts of all single signals. Based on the 5.3A vortex shaft reference in the proposed method, an integer number of segments is truncated, and the number of frequency shift cycles included in the synchronous segment is set accordingly. Autocorrelation is introduced, and then order tracking is performed based on the frequency shift, aiming to achieve comparison of performance indicators.

[0191] Figure 10 The comparison of the order spectrum of dynamic vortices is shown, in which Figure 10 (a) The noise floor energy is high in the order spectrum of the synchronous average signal, and the clarity of the harmonic order components is low. Figure 10 In (b), the anharmonic order component is dominant. Figure 11 The mid-index also reflects the shortcomings of the method, while at the same time it also reflects the advantage of autocorrelation in improving the significance of signal harmonic components.

[0192] Finally, through comparison with the other two methods, this invention draws statistically significant conclusions, proving the effectiveness of the proposed method in extracting harmonic noise components.

[0193] (1) Optimization effects of synchronous averaging and autocorrelation: Synchronous averaging and autocorrelation both have optimization effects in terms of harmonic significance and the proportion of harmonics in prominent spectral lines. This means that they can be used to extract and highlight important frequency components in the system, which is helpful for further signal analysis and fault detection.

[0194] (2) Advantages of synchronous averaging and autocorrelation: Synchronous averaging is advantageous in optimizing the proportion of harmonic frequencies of prominent spectral lines, while autocorrelation is advantageous in improving the significance of harmonic frequencies. Therefore, combining these two methods can achieve complementary advantages and process signals more comprehensively.

[0195] (3) Importance of Low-Frequency Interference Removal: In airborne and ground systems, removing low-frequency interference is crucial for improving the performance and accuracy of synchronization averaging. By removing low-frequency interference, unnecessary components in the noise signal can be effectively reduced, mitigating interference with phase alignment during synchronization and thus improving the effectiveness of synchronization averaging. This is critical for reliable signal processing.

[0196] (4) Application of order tracking and cross power spectrum: order tracking and discontinuous data segments can achieve data segment synchronization by performing phase compensation based on cross power spectrum based on the first rotation data, and have the ability to deal with emergency situations such as discontinuous acquisition and loss of some acquired data.

[0197] (5) Processing speed: The methods mentioned also have the advantage of high processing speed, which may be very important for real-time or high-frequency data processing in airborne systems. This ensures that the system responds promptly and processes data accurately.

[0198] The method provided by this invention can be applied to airborne and ground systems, helping to improve the efficiency and accuracy of data processing, analysis, and fault detection.

[0199] In summary, the above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for extracting harmonic noise components, characterized in that, Includes the following steps: Step 1: Under the same operating conditions, noise signals and rotational speed signals are collected separately in both airborne and ground systems. The time interval is set to be at least the time it takes for the rotor to be analyzed to rotate 2 revolutions. The noise signal and the speed signal are intercepted using the set time interval to obtain a series of short-time signals of equal length and corresponding short-time speed signals. Step 2: Process the series of short-time signals of equal duration and the corresponding short-time rotational speed signals to obtain frequency conversion data. Then, use the frequency conversion data to perform order tracking on the filtered signal and convert it into a stationary signal in the angular domain. Step 3: Perform synchronous averaging on the angular domain stationary signal to obtain a synchronous average signal; Step 4: The synchronous average signal is subjected to autocorrelation operation to obtain the autocorrelation signal, and then Fourier transform is performed to obtain the order spectrum, thus completing the extraction of harmonic noise components.

2. The method for extracting harmonic noise components as described in claim 1, characterized in that, In step one, under the same operating conditions in both airborne and ground systems, the noise signal is collected by a sound sensor; the rotational speed signal is collected by a rotational speed sensor.

3. The method for extracting harmonic noise components as described in claim 1, characterized in that, The specific steps for processing the series of short-time signals of equal duration and the corresponding short-time rotational speed signals to obtain frequency data are as follows: The series of short-time signals of equal length obtained in step one are xi(t), and the corresponding short-time rotational speed signal is n. i (t), where i is the sequence number; For each short-time speed signal n i (t) Find its average value mn i ; short-time signal x i (t) is transformed by Fourier FFT to obtain the spectrum x i (f) x i (f)=FFT(x i (t)) Where FFT(·) is the Fourier transform, f is the frequency, and the unit is Hz; Based on mn i and spectrum x i (f) Obtain frequency conversion data F(i) F(i)=nearmax(x i (f=mn i / 60)) nearmax(·) is used to calculate the spectrum x. i (f) Distance f = mn i The operation corresponding to the frequency value F(i) on the horizontal axis when the nearest maximum value is / 60, where F(i) is in Hz; First, observe the spectrum x. i (f), according to the spectrum x i (f) For the range of component concentration, set the upper limit frequency fh of the bandpass filter; secondly, set the lower limit frequency fl of the bandpass filter according to the sound field environment, fl is set empirically, and the filtering parameters [fl, fh] are applied to the short-time signal x. i (t) Perform bandpass filtering to obtain the filtered signal xf i (t); Define a rectangular window function w(f): Based on the given filter bandwidth, the spectrum x i (f) is multiplied by the rectangular window function w(f), and then the inverse Fourier transform is performed to obtain the corresponding filtered signal xf. i (t): xf i (f)=x i (f)·w(f) 。 4. The method for extracting harmonic noise components as described in claim 3, characterized in that, Step two: Process the series of short-time signals of equal duration and the corresponding short-time rotational speed signals to obtain frequency conversion data. Then, use the frequency conversion data to perform order tracking on the filtered signal, converting it into a stationary signal in the angular domain. Specifically: Based on the filtered signal xf i (t) corresponds to the conversion frequency F(i), and the conversion frequency F(i)×60 is copied and extended to the signal xf respectively. i (t) of the same length as in formula (7), then for xf i (t) Perform order tracking to obtain the corresponding angular domain signal ys i (r), where r is the number of revolutions; rpm i =(F(i)×60…F(i)×60…F(i)×60) 1×N （7） Where N is the signal xf i The data length of (t); Order tracking involves equal-angle sampling, which essentially means acquiring equal-angle moments. After acquiring these moments, interpolation and fitting on the original function yields the stationary signal in the angular domain, i.e., based on the shortest data length angular domain signal ys. i (r) The maximum integer revolutions L contained in all angular domain signals are uniformly truncated, and the truncated equal-length signal is defined as reys. i (n), where n represents the data points of the signal n = 1, 2, 3, ..., reys i (n) is the obtained angular domain stationary signal.

5. The method for extracting harmonic noise components as described in claim 4, characterized in that, Step three: Perform synchronous averaging on the angular domain stationary signal to obtain a synchronous average signal, which specifically includes the following steps: a. For angularly stationary signals reys i (n), the phase detection signal of its i-th segment is defined as: Where Fs is the resampling frequency set by the order tracking, and its value is equal to the number of sampling points in one revolution of the reference axis; b. Construct a reference cosine sequence of the same length as the phase detection signal. Then calculate the cross power spectrum P. i (k) and phase Phase is used for phase compensation: c. Convert the phase obtained in b into the number of sampling points. The expression is: The round(·) function performs rounding to the nearest integer. The phase difference between segments is eliminated by cyclic shifting. This cyclic shifting process can be represented as follows: in It is the i-th synchronization segment; d. Take the average value of all synchronization segments: Where xc is the final average synchronization signal, and Q is the number of synchronization segments.

6. A method for evaluating the extraction effect of a harmonic noise component extraction method, characterized in that, Harmonic noise component extraction is performed using a method for harmonic noise component extraction as described in any one of claims 1 to 5; The following three Fourier spectral domain quantitative indicators are used to measure the effectiveness of harmonic noise component extraction: harmonic significance (FFC), proportion of prominent spectral line harmonics (FFCnum), and comprehensive index (CL). Harmonic Significance (FFC): First, define the integer prominence multiplier Af. i Judgment criteria: Spectral lines in the order spectrum whose amplitude is greater than the threshold thr and whose corresponding order is an integer are defined as harmonic prominent spectral lines, where the spectral lines are harmonic prominent spectral lines in the spectrum. Where Af is the full spectrum amplitude, max(·) is the maximum value operation, and mean(·) is the average value operation. i To emphasize the amplitude of the harmonic order spectral lines, Af i The corresponding order is f i N * Represents positive integers; Secondly, the saliency range of the prominent harmonic spectral lines is defined as the open set (f). i -2,f i +2); Finally, the harmonic significance index value FFC is calculated according to the formula: Where f iw For the significance range, Af iw f iw For the corresponding spectral amplitude, sum(·) performs a summation operation; The proportion of harmonic frequencies of prominent spectral lines FFCnum: In the order spectrum, spectral lines whose amplitude exceeds the threshold thr defined in A are considered prominent spectral lines. When counting the number of prominent spectral lines, if the order values ​​of adjacent prominent spectral lines do not exceed 0.1, they are counted only once. Where FFCnum is the proportion of harmonic frequencies of prominent spectral lines, Af k To effectively highlight the spectral line amplitude, Af k The corresponding order is f k num(·) is a count operation; The comprehensive index CL is the sum of the two indices mentioned above: CL = FFC + FFCnum The comprehensive index CL is the sum of the two indices mentioned above, and aims to provide a balanced representation to characterize the overall performance of the studied method in terms of both the proportion of harmonic frequencies of prominent spectral lines and the significance of harmonic frequencies; through this comprehensive index, the effectiveness and feasibility of the method can be fully evaluated.

Images