An adaptive resonance feature identification method for a rotating blade tip timing signal

Abstract

The application discloses an adaptive resonance feature recognition method for a rotating blade tip timing signal and belongs to the technical field of structural vibration state monitoring. The application only needs a tip probe and does not need to manually set a threshold value, can provide a theoretical reference for analyzing the rotating blade vibration characteristics and adaptively marking a resonance region based on the tip timing test, and provides an efficient and reliable signal processing method for rotating blade vibration monitoring and fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of structural vibration state monitoring technology, and in particular to an adaptive resonance feature identification method for timing signals at the tip of rotating blades. Background Technology

[0002] Blades are critical components in major rotating equipment such as aero-engines, impeller pumps, and gas turbines. During operation, the blade's operating condition directly affects the safety and reliability of the entire mechanical system. In actual working conditions, blades are subjected to the combined effects of centrifugal force, unsteady aerodynamic forces, and alternating loads, resulting in a harsh working environment and making them one of the components with the highest failure rate in rotating machinery. Once fatigue cracks develop on a blade, if they are not detected and controlled in time, crack propagation will eventually lead to blade fracture and cause serious accidents. Therefore, conducting real-time vibration monitoring of rotating blades to ensure their operational stability and reliability is crucial.

[0003] During operation, rotating blades are excited by factors such as airflow wakes and rotor vibrations, resulting in synchronous vibration. If the fluid flow within the channel is unstable, or if blade tips rub against each other or foreign objects intrude, asynchronous vibration may occur, which can easily lead to abnormal blade vibration and fatigue failure. Therefore, extracting and identifying key information from the signals is of great significance for practical engineering. However, the analysis of this key information must be based on the identification of the resonance region. Traditional analysis methods rely on engineers manually judging the blade tip timing signal data, identifying the resonance region based on engineering experience, and extracting effective data to complete the vibration parameter identification. This process requires a high level of professional experience from the analysts, and the data processing cycle is long and inefficient. If online analysis of raw blade tip timing data and automatic identification of the resonance region can be achieved, the efficiency of blade vibration characteristic assessment can be significantly improved, early blade failures can be detected in a timely manner, and the risk of equipment accidents can be effectively reduced.

[0004] Only a few scholars have conducted relevant research on the automatic identification of resonance regions. Existing methods are mostly designed for synchronous or asynchronous signals and cannot adaptively identify resonance regions, exhibiting significant limitations.

[0005] Therefore, this invention proposes an adaptive resonance feature recognition method for timing signals at the tips of rotating blades. Summary of the Invention

[0006] This invention addresses the shortcomings of existing technologies by providing an adaptive resonance feature recognition method for timing signals at the tip of rotating blades. This method extracts the synchronous and asynchronous resonance components of the blade and proposes an automatic identification method for synchronous and asynchronous resonance regions. This provides a fast and effective means for online assessment of blade vibration status.

[0007] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: An adaptive resonance feature recognition method for timing signals at the tip of a rotating blade includes the following steps: Obtain the original blade tip timing signal, that is, collect the actual arrival time of the rotating blade during operation, and calculate the vibration displacement of the blade tip timing signal based on the difference between the actual arrival time and the theoretical arrival time. The acquired raw leaf tip timing signal is decomposed into synchronous and asynchronous components using the JMD method. Calculate the normalized cross-correlation function between each component and the original leaf tip timing signal, and extract the maximum cross-correlation coefficient; select the synchronous component based on the maximum cross-correlation coefficient, and select the asynchronous component by calculating the weighted scoring coefficient; For the selected synchronous and asynchronous components, an adaptive thresholding method based on local standard deviation is used to identify the resonance region.

[0008] Furthermore, by constructing a variational optimization model, the original leaf tip timing signal is decomposed into synchronous and asynchronous components.

[0009] Furthermore, the variational optimization model integrates the bandwidth constraints of the asynchronous components with the sparsity priors of the synchronous components.

[0010] Furthermore, the variational optimization model is constructed by weighted combination of broadband constraints, sparsity penalties, and reconstruction fidelity terms, and auxiliary variables are introduced. To address the non-differentiability of the penalty term, the constrained optimization of the variational optimization model is expressed as: ; In the formula, Synchronization Components Abbreviation; It is the center frequency; It is an asynchronous component; The sign of the partial derivative of the function with respect to time; For the asynchronous component bandwidth constraint, the weighting coefficients are... ; Let be the weighting coefficient for the sparsity penalty of the synchronization components, and ; To reconstruct the fidelity terms and ensure the consistency between the decomposed components and the original signal; This represents the original leaf tip timing signal; It is a natural imaginary number; The imaginary unit; For time; For the first The parameters of each signal; These are the control parameters for the penalty function; It is the square of the 2-norm of the vector; It is a non-convex penalty function. The objective function is denoted as .

[0011] Furthermore, the adaptive threshold method calculates the local standard deviation of each original leaf tip timing signal data point through a sliding window, and performs binarization processing in conjunction with the adaptive threshold. Then, it extracts the entire continuous oscillation region through continuity analysis to achieve a quantitative description of the characteristic parameters of each region. The judgment threshold is dynamically determined according to the statistical characteristics of the original leaf tip timing signal itself. Data points of the original leaf tip timing signal with local standard deviations exceeding the threshold are marked as oscillation points, and continuous oscillation points are connected to form a resonance region.

[0012] Furthermore, the step of connecting continuous oscillation points to form a resonance region specifically involves: performing continuity analysis on the labeled binarized sequence, merging data points with consecutive values ​​of 1 into a resonance region, and outputting the start position, end position, and peak position of each resonance region.

[0013] Furthermore, the statistical characteristics of the original leaf tip timing signal itself include standard deviation, peak-to-median ratio, and mean-to-median ratio; wherein, the basic adaptive coefficient is determined based on the peak-to-median ratio, and the more prominent the peak, the smaller the basic adaptive coefficient.

[0014] Furthermore, the vibration displacement of the blade tip timing signal is calculated as follows: ; In the formula, This indicates the circumferential vibration displacement of the blade tip; The instantaneous angular velocity of the rotor; The radius of the rotation trajectory of the blade tip; For the first blade in the first When the circle rotates, after the first... The time difference between the actual arrival time and the expected arrival time of each sensor.

[0015] Furthermore, the normalized cross-correlation function The formula for calculation is: ; In the formula, Represents component signals; This represents the original leaf tip timing signal; The number of delay points; For time.

[0016] Furthermore, a physical parameter S is proposed for selecting the asynchronous component, specifically through a weighted scoring coefficient. The calculation formula is as follows: ; In the formula, m is used to reflect the degree of fluctuation of the cross-correlation function. , For the cross-correlation root mean square, Represents the maximum cross-correlation number; For the first Delay penalty factor for each IMF component, ,in, The delay time corresponding to the maximum cross-correlation coefficient; The signal sampling frequency; The preset maximum delay points; Select Signals approaching 1 are used as synchronous signals, while the signal with the largest S value is selected as the asynchronous signal.

[0017] In summary, compared with the prior art, the beneficial effects of the above technical solution are: This invention requires only a single blade tip probe to complete signal acquisition, eliminating the need for multi-sensor deployment and debugging, effectively reducing hardware costs and on-site maintenance difficulty. Simultaneously, it adaptively adjusts the threshold based on the signal's inherent characteristics, eliminating the need for manual parameter setting and avoiding identification biases caused by human experience and poor adaptability to fixed thresholds. Combining JMD decomposition, component filtering, and local standard deviation algorithms, this invention can accurately distinguish and identify synchronous / asynchronous resonance regions, offering a high degree of automation and stable, reliable analysis results. It provides comprehensive theoretical reference for analyzing the vibration characteristics of rotating blades under timed blade tip testing and adaptively marking resonance regions. Furthermore, it offers an efficient, practical, and highly reliable signal processing solution for online monitoring and fault diagnosis of blade vibration in rotating equipment such as aero-engines and gas turbines. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the original timing signal at the leaf tip; Figure 2 This is a schematic diagram of the decomposed components of the leaf tip timing signal JMD; Figure 3 It is the result of calculating the maximum cross-correlation coefficient of each component; Figure 4 This is the result of calculating the weighted scoring coefficients for each component; Figure 5 This is the result of IMF1 resonance region identification; Figure 6 This is the process of identifying the local standard deviation of IMF1. Figure 7 It is the x-coordinate of the maximum value in the IMF1 resonance region; Figure 8 This is the result of IMF10 resonance region identification; Figure 9 This is the process of identifying the local standard deviation of IMF10. Figure 10 It is the x-coordinate of the maximum value in the IMF10 resonance region; Figure 11 This is a schematic diagram of the method according to an embodiment of the present invention. Detailed Implementation

[0019] The principles and features of the present invention are described below with reference to all the accompanying drawings. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0020] This invention discloses an adaptive resonance feature recognition method for timing signals at the tip of rotating blades. Specifically, it employs a blade tip timing raw data decomposition method based on Jump Plus Amplitude Modulation and Frequency Modulation Mode Decomposition (JMD), a synchronous and asynchronous signal selection method based on the maximum cross-correlation coefficient and weighted scoring coefficients, and a synchronous and asynchronous resonance region recognition method based on the local standard deviation of an adaptive threshold.

[0021] Reference Figures 1-11 An adaptive resonance feature recognition method for timing signals at the tip of a rotating blade includes the following steps: Obtain the original blade tip timing signal, that is, collect the actual arrival time of the rotating blade during operation, and calculate the vibration displacement of the blade tip timing signal based on the difference between the actual arrival time and the theoretical arrival time. The acquired raw leaf tip timing signal is decomposed into synchronous and asynchronous components using the JMD method. Calculate the normalized cross-correlation function between each component and the original leaf tip timing signal, and extract the maximum cross-correlation coefficient; select the synchronous component based on the maximum cross-correlation coefficient, and select the asynchronous component through the weighted scoring coefficient; For the selected synchronous and asynchronous components, an adaptive thresholding method based on local standard deviation is used to identify the resonance region.

[0022] The embodiments of the present invention require only one blade tip probe and do not require manual setting of thresholds. This invention can provide a theoretical reference for the analysis of the vibration characteristics of rotating blades based on blade tip timing tests and adaptive marking of resonance regions, and provide an efficient and reliable signal processing method for rotating blade vibration monitoring and fault diagnosis.

[0023] The above content will be explained in detail below: Step 1: Obtain the original blade tip timing signal, that is, collect the actual arrival time of the rotating blade during operation, and calculate the vibration displacement of the blade tip timing signal based on the difference between the actual arrival time and the theoretical arrival time.

[0024] Blade tip timing (BTT) is a non-contact technology for measuring the vibration of rotating blades. Multiple probes (mostly capacitive or fiber optic) are mounted on a stationary casing to monitor the arrival time of the blade at each sensor. When a high-speed rotating blade passes the measurement position of a BTT probe, the sensor accurately captures the arrival time. Ideally, the blade arrival time would follow a periodic distribution. However, in actual operating conditions, blade vibration causes a deviation between the actual arrival time and the expected arrival time under ideal, vibration-free conditions. This time difference... It directly reflects the vibration characteristics of the blades, among which For the first The first leaf in the When the circle rotates, after the first... The time difference between the actual and expected arrival times of each sensor For the actual situation, the first The time it takes for each blade to reach the sensor For the theoretical first The time it takes for each blade to reach the sensor.

[0025] Combining the instantaneous angular velocity of the blade rotor and the blade tip rotation radius, the time deviation can be converted into the circumferential vibration displacement of the blade through geometric relationships. Therefore, the blade's displacement in the first... The expression for the in-circuit vibration displacement is:

[0026] In the formula, This indicates the circumferential vibration displacement of the blade tip; The instantaneous angular velocity of the rotor; The radius of the rotation trajectory of the blade tip; For the first blade in the first When the circle rotates, after the first... The time difference between the actual arrival time and the expected arrival time of each sensor.

[0027] Step 2: Decompose the acquired raw leaf tip timing signal into synchronous and asynchronous components using the JMD method.

[0028] JMD is a novel decomposition algorithm for non-stationary signals. It integrates tools and methods from traditional amplitude modulation (AM) and frequency modulation (FM) signal decomposition and jump extraction concepts to extract jump components and inherent oscillation components from multivariable signals. Therefore, by combining the principles of the JMD algorithm with the characteristics of the leaf tip timing signal, the aim is to separate the asynchronous component (AM-FM) and the synchronous component (Jump) in the signal. These are also referred to as synchronous resonance signals, asynchronous vibration signals, or synchronous signal components and asynchronous signal components.

[0029] The JMD algorithm integrates the bandwidth constraints of asynchronous components and the sparsity priors of synchronous components by constructing a variational optimization model, achieving joint optimization extraction of the two types of components, and is suitable for single-channel and multi-channel signal processing.

[0030] This algorithm employs a single-stage optimization framework, avoiding the parameter tuning complexity of multi-stage methods; it accurately characterizes the discontinuity of the synchronous component through a non-convex penalty term; and it utilizes analytic signal transformation to achieve bandwidth minimization constraints for the asynchronous component, ensuring the purity of the oscillation mode. Therefore, the original blade tip timing signal can be represented by the summation of the following components:

[0031] in, For synchronization components; Represents asynchronous components; Indicates trend components; This represents the original leaf tip timing signal.

[0032] Solving this optimization problem using the Alternating Direction Multiplier Method (ADMM) enables iterative optimization extraction of synchronous and asynchronous components. The complete variational optimization model of JMD is constructed through a weighted combination of bandwidth constraints, sparsity penalties, and reconstruction fidelity terms, introducing auxiliary variables. To handle the non-differentiability of the penalty term, the constrained optimization form is as follows:

[0033] Synchronization Components Abbreviation; It is the center frequency; It is an asynchronous component; is the sign of the partial derivative of the function with respect to time; α represents the weighting coefficient of the asynchronous component bandwidth constraint, and β is the weighting coefficient for the sparsity penalty of the synchronization component, and ; To reconstruct the fidelity terms and ensure the consistency between the decomposed components and the original signal; It is a natural imaginary number; The imaginary unit; For time; For the first The parameters of each signal; These are the control parameters for the penalty function; It is the square of the 2-norm of the vector; It is a non-convex penalty function.

[0034] The constrained optimization is transformed into an unconstrained augmented Lagrangian function using the Lagrange multiplier method. :

[0035] in, To change the non-convexity parameter; It is a penalty scalar parameter; Represents the dual variable associated with the constraint.

[0036] The alternating direction multiplier method (ADMM) is then used to solve the above formula, decomposing the augmented Lagrangian function into multiple subproblems and iteratively solving each variable.

[0037] right Solving the minimization optimization problem, fixed ,in Synchronization signal component Abbreviation; for Abbreviation. Extract only the terms that are related to them, and then perform a Fourier transform to solve. By performing partial derivative calculations and algebraic transformations, the following iterative formula is obtained:

[0038]

[0039] in, Indicates Fourier transform, Indicates the first The next iteration.

[0040] for Minimization: The suboptimal problem to be solved can be derived from formula (4), converted into a discrete form, and solved to obtain the solution. The steps of the iterative equation.

[0041]

[0042] in, Defined as the first derivative matrix:

[0043] in, Represents the set of real numbers. This represents the dimension of the matrix, with 100 rows. The number of columns is P.

[0044] Next, by solving the formula Take about The partial derivatives of the vector, and the partial derivatives of the vector. To perform zero initialization, you can obtain the following: The Next iteration:

[0045] in, for Transpose of a matrix. To and An identity matrix of the same order is used to ensure matrix invertibility and enhance the stability of iterative solutions.

[0046] Auxiliary variables The iterations obtain a closed-form solution by solving a one-dimensional optimization problem:

[0047] in, , , For the first matrix One component; It is a penalty scalar parameter; it should be noted that, , These parameters have no specific meaning; they are simply set to facilitate the description of formula (10). The iterations will follow:

[0048] The detailed steps of the JMD algorithm for decomposing signals are shown in Table 1.

[0049] Table 1. Steps for decomposing the leaf tip resonance signal

[0050] Step 3: Calculate the normalized cross-correlation function between each component and the original leaf tip timing signal, and extract the maximum cross-correlation coefficient; select the synchronous component based on the maximum cross-correlation coefficient, and select the asynchronous component based on the weighted scoring coefficient.

[0051] This embodiment proposes a method for extracting synchronous and asynchronous signal components based on time-domain cross-correlation. Using the original blade tip timing signal as a reference, the method selects synchronous and asynchronous signals by quantifying the time-domain cross-correlation between each decomposed vibration component and the original blade tip timing signal. Then, a physical parameter that can effectively characterize the signal features is proposed, and the asynchronous signal is selected using this parameter.

[0052] Maximum correlation coefficient It is the maximum value of the normalized cross-correlation function, which can represent the highest degree of matching under the globally optimal delay, characterizing waveform similarity. For each component... Define it as a relative to the original leaf tip timing signal. The formula for calculating the normalized cross-correlation function is:

[0053] in, This represents the delay points.

[0054] Normalized cross-correlation functions were calculated for the original leaf tip timing signal and each component signal, and the maximum cross-correlation coefficient was selected. Parameters for the synchronization signal are selected based on calculation results, and are chosen to be close to 1. As a synchronization signal.

[0055] The maximum cross-correlation coefficient is sufficient for the effective selection of synchronous signals; all other signals fall into the category of asynchronous signals. Furthermore, signals with a maximum cross-correlation coefficient approaching 0 may not necessarily have superior signal recognition performance in later stages. Therefore, it is necessary to construct a new physical parameter S to enable the selection of asynchronous signals.

[0056] The root mean square of the cross-correlation can measure the dispersion of the cross-correlation sequence and reflect the correlation energy distribution. Based on this parameter, a physical parameter is proposed. The asynchronous signal is selected by using a weighted scoring coefficient, and the calculation formula is as follows:

[0057] in, It can be used to reflect the degree of fluctuation of the cross-correlation function; the larger the value, the smoother the signal and the less noise interference. For the first The delay penalty factor for each IMF component; where The delay time corresponding to the maximum cross-correlation coefficient; The signal sampling frequency; This is the preset maximum delay point.

[0058] In summary, the calculation result is selected to be close to 1. As a synchronization signal, select The largest signal is used as an asynchronous signal.

[0059] Step 4: For the selected synchronous and asynchronous components, the adaptive thresholding method based on local standard deviation is used to identify the resonance region.

[0060] To identify the resonance regions of synchronous and asynchronous signals and the location of peaks within these regions, this invention proposes an adaptive threshold identification method based on local standard deviation. This algorithm adaptively selects a threshold based entirely on the characteristics of the signal itself and identifies the resonance regions and peaks.

[0061] Asynchronous regions typically exhibit drastic local amplitude fluctuations, with their local standard deviations significantly higher than those in stable regions. Therefore, a sliding window is used to calculate the local standard deviation, which is then combined with an adaptive threshold for binarization. Continuity analysis is then used to extract the entire continuous oscillation region, ultimately achieving a quantitative description of the characteristic parameters of each region.

[0062] The algorithm uses a sliding window method to quantize and calculate the local standard deviation of each raw leaf tip timing signal data point, thereby determining whether the data point is an oscillation point. The choice of window size directly affects the calculation result of the local standard deviation. If the window is too small, it will be too sensitive to noise and easily misjudge random fluctuations as oscillations; if the window is too large, it will reduce the spatial resolution of the detection and may cause adjacent oscillation regions to merge. To resolve this contradiction, this algorithm uses an active window size. :

[0063] Here, 40 is a suitable size selected based on the number of signal data points to be identified. The size can be adjusted appropriately according to the amount of data and the desired effect. N is the total number of data points.

[0064] This strategy ensures sufficient statistical stability of the window size in long signal sequences while avoiding edge effects caused by an excessively large window size relative to the signal length. Therefore, for the... For each data point, define its neighborhood index set. ,as follows:

[0065] Let the number of points in the neighborhood be . The mean of the amplitude within the neighborhood As shown in the following formula:

[0066] In summary, the local standard deviation is as follows:

[0067] Obtaining the local standard deviation sequence Then, the asynchronous point indicator sequence is obtained through threshold comparison. Binarization processing defines the asynchronous point indicator variable: when... The value is 1 when it is greater than the adaptive threshold, and 0 otherwise.

[0068] This algorithm proposes an adaptive threshold setting mechanism. Using the standard deviation of the signal as a benchmark, it dynamically determines the scaling factor through a series of statistical features, ultimately forming an adaptive threshold proportional to the signal fluctuation. The key point is that the local standard deviation in the asynchronous region should be slightly larger than the overall standard deviation of the signal, but the degree of this larger deviation needs to be adjusted based on the signal's peak and distribution characteristics.

[0069] Based on this characteristic, select the maximum absolute value. Median of absolute value The ratio is used as a parameter of the adaptive threshold, and is defined as follows: The calculation formula is as follows:

[0070] in, It is a very small positive number, used to prevent division by zero errors.

[0071] This ratio quantifies the prominence of signal spikes relative to typical amplitudes: The larger the value, the more significant the peaks in the signal; if it approaches 1, it indicates a more uniform signal amplitude distribution. According to... The range of values ​​is used to classify signals and assign different basic adaptive coefficients to each. As shown in Table 2.

[0072] Table 2. Classification and Selection of Basic Coefficients

[0073] For signals with obvious spikes, the asynchronous region usually exhibits continuous fluctuations rather than isolated spikes, so a smaller threshold is needed to ensure the integrity of the oscillation region; for signals with uniform amplitude distribution, the amplitude of random noise is relatively small, and a larger threshold can effectively filter out noise interference.

[0074] Define the ratio of mean to median As shown in the following formula:

[0075] This ratio reflects the degree of skewness in the distribution of the absolute value of the signal. When the signal distribution is perfectly symmetrical, the means are approximately equal. When the signal exhibits a right-skewed distribution (i.e., contains some large amplitude points), the mean is greater than the median. .like This indicates a significant signal skew, meaning that large amplitude points have a noticeable impact on the mean. For significantly skewed signals, the unevenness of their amplitude distribution may amplify the standard deviation in local regions. To avoid over-segmentation of these regions, the algorithm introduces a skew adjustment factor:

[0076] This adjustment increases the adaptive coefficient by 10%, thereby raising the detection threshold and avoiding false detections caused by skewed distribution.

[0077] Based on the aforementioned adaptive coefficients, the final standard deviation threshold is... for:

[0078] Multiplying the overall standard deviation of the signal by an adaptive coefficient yields a critical value for judging local standard deviations. When the local standard deviation at a point exceeds this critical value, the point is determined to belong to an oscillation region. Finally, A continuous series of points equal to 1 forms a resonance region.

[0079] The content of this invention will be further explained below with reference to specific examples: Step 1: Timed data acquisition of leaf tips.

[0080] In a specific embodiment, in the blade tip timing signal monitoring system, four blade tip timing sensors (TIP0, TIP1, TIP2, and TIP3) are installed along the circumference of the blade tip, with their relative installation angles to TIP0 being respectively... The vibration displacement of the blade tip was acquired using the key phase method. During the experiment, several air nozzles were installed on the side of the rotating disk, and an air pump was used to excite the rotating blades, achieving synchronous vibration excitation during blade rotation. The blade rotation frequency range was set to 30-50Hz. Based on the BTT test principle, the vibration displacement of different blades was obtained, and the data from the TIP0 sensor was selected as the source signal for subsequent processing. To achieve asynchronous resonance, the blades were subjected to rubbing excitation during the experiment. An array of blade tip timing signal data was obtained based on the blade tip timing test principle.

[0081] Step 2: Decomposition of blade tip vibration signal.

[0082] Select one set of data for JMD signal decomposition and set the parameters: ; ; ; ; The result is as follows Figure 2 As shown.

[0083] Step 3: Selection of synchronous and asynchronous signal components.

[0084] Synchronous and asynchronous signal components were selected from the IMF1-IMF10 obtained by the above decomposition. The maximum cross-correlation coefficient was calculated by comparing each of the 10 components with the original signal. The calculation results are as follows: Figure 3 As shown, weighted scoring coefficients are then applied. The calculation results are as follows: Figure 4 As shown, the signal with the largest cross-correlation coefficient close to 1 is selected as the synchronization signal, i.e., IMF1. The largest is the asynchronous signal, namely IMF10.

[0085] Step 4: Resonance region identification based on adaptive threshold.

[0086] Based on the above results, adaptive threshold identification of resonance regions was performed on IMF1 and IMF10, and the results are as follows: Figure 5-10 As shown in the figure. The results show that the invention has a good effect on the identification of resonance regions for both synchronous and asynchronous resonance signals.

[0087] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An adaptive resonance feature recognition method for timing signals at the tip of a rotating blade, characterized in that, Includes the following steps: Obtain the original blade tip timing signal, that is, collect the actual arrival time of the rotating blade during operation, and calculate the vibration displacement of the blade tip timing signal based on the difference between the actual arrival time and the theoretical arrival time. The acquired raw leaf tip timing signal is decomposed into synchronous and asynchronous components using the JMD method. Calculate the normalized cross-correlation function between each component and the original leaf tip timing signal, and extract the maximum cross-correlation coefficient; Synchronous components are selected based on the maximum cross-correlation coefficient, and asynchronous components are selected by calculating weighted scoring coefficients. For the selected synchronous and asynchronous components, an adaptive thresholding method based on local standard deviation is used to identify the resonance region.

2. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 1, characterized in that: By constructing a variational optimization model, the original leaf tip timing signal is decomposed into synchronous and asynchronous components.

3. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 2, characterized in that: The variational optimization model integrates the bandwidth constraints of asynchronous components with the sparsity priors of synchronous components.

4. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 3, characterized in that, The variational optimization model is constructed by weighted combination of broadband constraints, sparsity penalties, and reconstruction fidelity terms, and auxiliary variables are introduced. To address the non-differentiability of the penalty term, the constrained optimization of the variational optimization model is expressed as: ； In the formula, Synchronization Components Abbreviation; It is the center frequency; It is an asynchronous component; The sign of the partial derivative of the function with respect to time; For the asynchronous component bandwidth constraint, the weighting coefficients are... ; Let be the weighting coefficient for the sparsity penalty of the synchronization components, and ; To reconstruct the fidelity terms and ensure the consistency between the decomposed components and the original signal; This represents the original leaf tip timing signal; It is a natural imaginary number; The imaginary unit; For time; For the first The parameters of each signal; These are the control parameters for the penalty function; It is the square of the 2-norm of the vector; It is a non-convex penalty function. The objective function is denoted as .

5. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 1, characterized in that: The adaptive threshold method calculates the local standard deviation of each original leaf tip timing signal data point through a sliding window, and performs binarization processing in conjunction with the adaptive threshold. Then, it extracts the entire continuous oscillation region through continuity analysis to achieve a quantitative description of the characteristic parameters of each region. The judgment threshold is dynamically determined based on the statistical characteristics of the original leaf tip timing signal itself. Data points of the original leaf tip timing signal with a local standard deviation exceeding the threshold are marked as oscillation points, and continuous oscillation points are connected to form a resonance region.

6. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 5, characterized in that, The step of connecting continuous oscillation points to form a resonance region specifically involves: performing continuity analysis on the labeled binarized sequence, merging data points with consecutive values ​​of 1 into a resonance region, and outputting the start position, end position, and peak position of each resonance region.

7. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 5, characterized in that, The statistical characteristics of the original leaf tip timing signal include standard deviation, peak-to-median ratio, and mean-to-median ratio; among which, the basic adaptive coefficient is determined based on the peak-to-median ratio, and the more prominent the peak, the smaller the basic adaptive coefficient.

8. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 1, characterized in that, The vibration displacement of the blade tip timing signal is calculated as follows: ； In the formula, This indicates the circumferential vibration displacement of the blade tip; The instantaneous angular velocity of the rotor; The radius of the rotation trajectory of the blade tip; For the first blade in the first When the circle rotates, after the first... The time difference between the actual arrival time and the expected arrival time of each sensor.

9. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 1, characterized in that: The normalized cross-correlation function The formula for calculation is: ； In the formula, Represents component signals; This represents the original leaf tip timing signal; The number of delay points; For time.

10. The adaptive resonance feature recognition method for timing signals at the tip of a rotating blade according to claim 9, characterized in that, The physical parameter S is proposed to select the asynchronous component, that is, to select the asynchronous component through a weighted scoring coefficient. The calculation formula is as follows: ； In the formula, m Used to reflect the degree of fluctuation of the cross-correlation function. , For the cross-correlation root mean square, Represents the maximum cross-correlation number; For the first Delay penalty factor for each IMF component, ,in, The delay time corresponding to the maximum cross-correlation coefficient; The signal sampling frequency; The preset maximum delay points; Select Signals approaching 1 are used as synchronous signals, while the signal with the largest S value is selected as the asynchronous signal.

Images