Blade vibration signal noise reduction method based on engine state division wavelet interval

Abstract

The invention discloses a blade vibration signal noise reduction method based on engine state division wavelet intervals. The method comprises the following steps: collecting an original noisy signal of an engine based on a dynamic signal analyzer; performing wavelet decomposition on the original noisy signal; carrying out one-dimensional wavelet reconstruction on the noisy signal after wavelet decomposition; and obtaining the denoised signal. By means of the method, it can be guaranteed that the filtered phase does not generate obvious distortion, meanwhile, noise contained in the vibration signals can be filtered out, noise elimination is achieved, the signal-to-noise ratio of the signals is increased, and data with the high signal-to-noise ratio is provided for engine dynamic stress spectrum analysis.

Description

Technical Field

[0001] The invention belongs to the field of aero-engine blade dynamic stress test data preprocessing, and in particular relates to a blade vibration signal denoising method based on engine state division of wavelet intervals. Background Art

[0002] The dynamic frequency and dynamic stress test method of aircraft engine blades is carried out using the "strain gauge + signal telemetry transmission" method. Strain gauges are installed at the blade measuring points, and the signals are transmitted to the data acquisition system using measuring leads. The lead path is long, and the number of blades is large. The blade working conditions are complex, and various components and interferences are easily coupled in the measured signals, which is very unfavorable for the extraction of dynamic stress signals. Therefore, data denoising, filtering, detrending and other preprocessing must be performed before formal data analysis.

[0003] When measuring the dynamic stress of aircraft engine blades, it is necessary to quickly and accurately analyze and process the measured vibration signals to obtain key and accurate vibration data, obtain the dynamic frequency-speed diagram (Campbell diagram) to determine the stress-speed characteristics of the blades in the operating speed range, determine whether there is any dangerous resonance in the blades within the engine operating speed range, and complete the verification of the theoretical calculated strength and the actual measured strength of the blades. At the same time, it is necessary to reduce the large amount of time spent on data processing to ensure that the test data results are more accurate, convenient and efficient.

[0004] However, the commonly used classical filter IIR digital signal filtering will cause signal phase distortion; FIR digital filters have more orders than IIR filters, require more storage units, and have a large amount of calculations, and have poor filtering amplitude-frequency characteristics, and cannot filter out noise completely.

[0005] Therefore, there is an urgent need for a method to denoise vibration signals, which must not only ensure that the phase after filtering does not produce obvious distortion, but also filter out the noise contained in the vibration signal, achieve noise removal, improve the signal-to-noise ratio of the signal, and provide high signal-to-noise ratio data for the next step of dynamic stress test data spectrum analysis. Summary of the invention

[0006] In order to solve the above technical problems, the present application provides a method for reducing the noise of blade vibration signals by dividing wavelet intervals based on engine state, comprising the following steps:

[0007] The original noisy signal of the engine is collected based on the dynamic signal analyzer;

[0008] Perform wavelet decomposition on the original noisy signal;

[0009] Perform one-dimensional wavelet reconstruction on the noisy signal after wavelet decomposition;

[0010] Get the signal after noise reduction.

[0011] Furthermore, the original noisy signal is an engine speed signal and a blade vibration signal.

[0012] Furthermore, before the wavelet decomposition, the following steps are performed:

[0013] Decompose the original noisy signal into different levels to obtain noisy signals at different levels;

[0014] Segmenting the engine state according to the change of the engine speed signal to obtain different engine states;

[0015] Selecting corresponding wavelet functions for noisy signals at different levels according to the engine state;

[0016] The divided different engine states are used as division intervals;

[0017] A threshold function is established for each interval, and the threshold is automatically selected based on the statistical characteristics of the signal.

[0018] Furthermore, the wavelet function includes sym2, sym4, sym8, DB2, DB4, DB8, coif3 or coif5.

[0019] Furthermore, the threshold is determined using the Birge-Massart strategy, and the calculation method includes the following steps:

[0020] Get the number of decomposition levels;

[0021] If the decomposition level is greater than or equal to the specified decomposition level n, all wavelet coefficients are retained;

[0022] If the decomposition level is less than the specified decomposition level m, and 1≤m≤n, retain pm wavelet coefficients with the largest absolute value, pm is determined by the formula pm=M(n+2-m)a, where: M and a are empirical coefficients, M=L(1) is the length of the wavelet coefficients after the first layer of decomposition, and a is used as the noise reduction coefficient a=3.

[0023] Furthermore, one-dimensional wavelet reconstruction is performed according to a threshold function and a partition interval.

[0024] Furthermore, the one-dimensional wavelet reconstruction includes: selecting a suitable threshold value for each interval through an established threshold function, retaining wavelet coefficients greater than the threshold value, and setting wavelet coefficients less than the threshold value to zero to complete denoising.

[0025] Furthermore, after obtaining the denoised signal, the method further includes measuring the similarity of the denoised signal;

[0026] The similarity of the denoised signal is measured by comparing the ratio perc of the denoised result and the original signal, the standard deviation err of the denoised result and the original signal, and the original noisy signal.

[0027] Furthermore, the calculation formula of the ratio perc of the noise reduction result to the original signal is:

[0028]

[0029] Among them, d i is the result signal after noise reduction, s i is the original signal.

[0030] Furthermore, the calculation formula for the standard deviation err between the noise reduction result and the original signal is:

[0031]

[0032] Among them, d i is the result signal after noise reduction, s i is the original signal.

[0033] The beneficial effect of the present invention is that the noisy signal is processed in segments according to the preset speed change rate, and the wavelet filter implemented by adaptively using different thresholds is automatically and quickly noise-reduced according to different working states of the engine to obtain blade vibration data with a higher signal-to-noise ratio. This method can not only ensure that the phase after filtering does not produce obvious distortion, but also filter out the noise contained in the vibration signal, realize noise removal, improve the signal-to-noise ratio of the signal, and provide high signal-to-noise ratio data for engine dynamic stress spectrum analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 is a flowchart of blade vibration signal denoising based on engine state division wavelet interval according to an embodiment of the present invention;

[0035] Figure 2 It is a segmentation diagram of different engine states according to an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The technical solution of the present invention is further described below, but the scope of protection claimed is not limited to the description.

[0037] The embodiment of the present invention provides a method for reducing the noise of blade vibration signals based on the wavelet interval division of the engine state, and uses different interval correlation thresholds and adaptive wavelet decomposition to automatically and quickly reduce the noise of non-stationary vibration signals of engine blades at various stages. The method includes the following steps:

[0038] Step S10: collecting the original noisy signal of the engine based on the dynamic signal analyzer.

[0039] The original noisy signal includes the engine speed signal and the blade vibration signal.

[0040] Step S20: performing wavelet decomposition on the original noisy signal. Before the wavelet decomposition, the following steps are performed:

[0041] Step S21: Decompose the original noisy signal into levels to obtain noisy signals of different levels.

[0042] When using wavelet denoising in Matlab, determining the decomposition level is a key step, which directly affects the effect of denoising and computational efficiency. Factors and methods that need to be considered when determining the decomposition level are:

[0043] Considerations:

[0044] (1) Signal characteristics: Different signals have different frequency characteristics and noise levels, so it is necessary to select the appropriate decomposition level based on the characteristics of the signal. Generally speaking, signals with more high-frequency noise may require more decomposition levels to effectively remove the noise.

[0045] (2) Balance between noise reduction effect and computational complexity: As the number of decomposition levels increases, the noise reduction effect may improve, but the computational complexity will also increase. Therefore, it is necessary to find a balance between noise reduction effect and computational complexity.

[0046] (3) Selection of wavelet basis: Different wavelet bases have different properties, such as regularity, symmetry, vanishing moment, etc. These properties will affect the effect of wavelet decomposition. Therefore, when selecting the decomposition level, the characteristics of the selected wavelet basis also need to be considered.

[0047] Determine method:

[0048] (1) Empirical method: Select the decomposition level based on experience or recommendations in the literature.

[0049] (2) Experimental method: By testing different decomposition levels multiple times, observe the noise reduction effect and the change in the amount of calculation, and then select the most appropriate decomposition level. You can choose 4-5 levels to perform calculations and observe the settlement results.

[0050] (3) Automatic selection method: Use functions or toolboxes in Matlab to automatically select the decomposition level. For example, you can use the ddencmp function to obtain the default threshold for denoising or compression, and select the appropriate level by observing the denoising effect at different decomposition levels.

[0051] Step S22: Segment the engine state according to the change of the engine speed signal to obtain different engine states.

[0052] The change of the speed signal is preset, and the different engine states, from the process of engine starting to stopping, are divided into 5 states A to E, specifically divided as follows: Figure 2 .

[0053] Step S23: Select corresponding wavelet functions for noisy signals of different levels according to the engine state.

[0054] The wavelet function includes sym2, sym4, sym8, DB2, DB4, DB8, coif3 or coif5

[0055] Step S24: The divided different engine states are used as divided intervals.

[0056] Step S25: Establish a threshold function for each interval and automatically select the threshold according to the statistical characteristics of the signal.

[0057] The threshold is determined using the Birge-Massart strategy, and the calculation method includes the following steps:

[0058] Get the number of decomposition levels;

[0059] If the decomposition level is greater than or equal to the specified decomposition level n, all wavelet coefficients are retained;

[0060] If the decomposition level is less than the specified decomposition level m, and 1≤m≤n, retain pm wavelet coefficients with the largest absolute value, pm is determined by the formula pm=M(n+2-m)a, where: M and a are empirical coefficients, M=L(1) is the length of the wavelet coefficients after the first layer of decomposition, and a is used as the noise reduction coefficient a=3.

[0061] Step S30: Perform one-dimensional wavelet reconstruction on the noisy signal after wavelet decomposition. This stage specifically includes the following operations:

[0062] One-dimensional wavelet reconstruction is performed based on the threshold function and the partition interval.

[0063] The specific reconstruction denoising method is: through the established threshold function, a suitable threshold is selected for each interval, the wavelet coefficients greater than the threshold are retained, and the wavelet coefficients less than the threshold are set to zero to complete the denoising.

[0064] Step S40: Obtain the noise-reduced signal.

[0065] Step S41: After one-dimensional wavelet reconstruction, obtain the denoised signal.

[0066] Step S42: after obtaining the denoised signal, the method further includes measuring the similarity of the denoised signal.

[0067] The similarity of the denoised signal is measured by comparing the ratio perc of the denoised result and the original signal, the standard deviation err of the denoised result and the original signal, and the original noisy signal.

[0068] The calculation formula of the ratio perc of the noise reduction result to the original signal is:

[0069]

[0070] The calculation formula of the standard deviation err between the noise reduction result and the original signal is:

[0071]

[0072] Among them, d i is the result signal after noise reduction, s i is the original signal.

[0073] The wavelet denoising principle in this embodiment is as follows: after the signal is decomposed by wavelet, the wavelet coefficient of the effective signal is larger, and the wavelet coefficient of the noise is smaller, and the wavelet coefficient of the noise is smaller than the wavelet coefficient of the effective signal. By selecting a suitable threshold, the wavelet coefficients greater than the threshold are considered to be generated by the signal and should be retained, while those less than the threshold are considered to be generated by noise and are set to zero to achieve the purpose of denoising.

[0074] There are three basic problems in wavelet threshold denoising: the selection of wavelet basis, the selection of threshold, and the selection of threshold function.

[0075] (1) Selection of wavelet basis: Usually we hope that the selected wavelet satisfies the following conditions: orthogonality, high vanishing moment, compact support, symmetry or antisymmetry.

[0076] (2) Threshold selection: An important factor that directly affects the denoising effect is the selection of the threshold. Different threshold selections will have different denoising effects. Currently, there are mainly general thresholds (VisuShrink), SureShrink thresholds, Minimax thresholds, BayesShrink thresholds, etc.

[0077] (3) Selection of threshold function: The threshold function is the rule for modifying the wavelet coefficients. Different threshold functions reflect different strategies for processing wavelet coefficients. There are two most commonly used threshold functions: one is the hard threshold function and the other is the soft threshold function. In addition, there is also a Garrote function that is between the soft and hard threshold functions.

[0078] The present invention provides a method for reducing the noise of blade vibration signals related to wavelet intervals based on engine state division, performs segmented processing according to a preset speed change rate, and adaptively adopts wavelet filters implemented with different thresholds to automatically and quickly reduce the noise according to different engine working states without changing the phase of the signal, so as to obtain blade vibration data with a higher signal-to-noise ratio.

[0079] The above disclosure is only a specific embodiment of the present invention, but the present invention is not limited thereto, and any changes that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A method for reducing the noise of blade vibration signals by dividing wavelet intervals based on engine state, characterized in that: The following steps are involved: The original noisy signal of the engine is collected based on the dynamic signal analyzer; Perform wavelet decomposition on the original noisy signal; Perform one-dimensional wavelet reconstruction on the noisy signal after wavelet decomposition; Get the signal after noise reduction.

2. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals according to claim 1, characterized in that: The original noisy signal includes an engine speed signal and a blade vibration signal.

3. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 2, characterized in that: Before the wavelet decomposition, the following steps are performed: Decompose the original noisy signal into different levels to obtain noisy signals at different levels; Segmenting the engine state according to the change of the engine speed signal to obtain different engine states; Selecting corresponding wavelet functions for noisy signals at different levels according to the engine state; The divided different engine states are used as division intervals; A threshold function is established for each interval, and the threshold is automatically selected based on the statistical characteristics of the signal.

4. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 3, characterized in that: The wavelet function includes sym2, sym4, sym8, DB2, DB4, DB8, coif3 or coif5.

5. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 3, characterized in that: The threshold is determined using the Birge-Massart strategy, and the calculation method includes the following steps: Get the number of decomposition levels; If the decomposition level is greater than or equal to the specified decomposition level n, all wavelet coefficients are retained; If the decomposition level is less than the specified decomposition level m, and 1≤m≤n, retain the p with the largest absolute value m wavelet coefficients, p m According to the formula p m =M(n+2-m) a Determine, where: M and a are empirical coefficients, M=L(1) is the length of the wavelet coefficients after the first layer of decomposition, and a is used as the noise reduction coefficient a=3.

6. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals according to claim 1, characterized in that: The one-dimensional wavelet reconstruction is performed according to a threshold function and a partition interval.

7. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 6, characterized in that: The one-dimensional wavelet reconstruction includes: selecting a suitable threshold value for each interval through an established threshold function, retaining wavelet coefficients greater than the threshold value, and setting wavelet coefficients less than the threshold value to zero to complete denoising.

8. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 1, characterized in that: After obtaining the noise-reduced signal, the method further includes measuring the similarity of the noise-reduced signal; The similarity of the denoised signal is measured by comparing the ratio perc of the denoised result and the original signal, the standard deviation err of the denoised result and the original signal, and the original noisy signal.

9. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 8, characterized in that: The calculation formula of the ratio perc of the noise reduction result to the original signal is: Among them, d i is the result signal after noise reduction, s i is the original signal.

10. The method for reducing denoising of blade vibration signals based on engine state division of wavelet intervals as claimed in claim 8, characterized in that: The calculation formula of the standard deviation err between the noise reduction result and the original signal is: Among them, d i is the result signal after noise reduction, s i is the original signal.

Images