A signal purification method for a waveguide rod online monitoring system

Abstract

The waveguide rod on-line monitoring system is a reliable long-term measurement technology for inner wall thinning of high-temperature structure, but there are many environmental interference factors in industrial field, and the collected signals often mix noise, which brings error to the monitoring result. The application discloses a waveguide rod on-line monitoring system signal purification method, which surrounds the 5-period sine pulse ultrasonic wave signal modulated by Hanning window received in the waveguide rod on-line monitoring system, adopts db8 wavelet as a base function, and carries out decomposition with a scale of 7 layers. The threshold value is determined by using the unbiased risk estimation method based on Stein, the re-adjustment threshold value is fixed based on the product (sln) of the obtained threshold value and the fixed first layer decomposition noise variance, the soft threshold value method is used to determine the coefficients of each layer, and finally the signal is reconstructed. The threshold value processing method based on wavelet denoising can effectively filter out the noise components in the original ultrasonic wave signals collected under the actual working condition, achieve a better signal purification effect, and improve the measurement accuracy.

Description

Technical Field

[0001] This invention relates to signal processing methods for ultrasonic monitoring, and particularly to a method for purifying 5-cycle sinusoidal pulse ultrasonic signals modulated by a Hanning window in an online waveguide rod monitoring system. It belongs to the field of ultrasonic non-destructive testing. Background Technology

[0002] The waveguide rod online monitoring system (ZL201810350599.X) can be used in petroleum, chemical, and power industries to measure the damage propagation processes of critical pressure-bearing equipment, such as erosion and corrosion thinning. However, industrial sites often contain a large amount of environmental noise, such as machine vibration, liquid flow, electromagnetic interference, and coarse grain scattering. During monitoring, the presence of these noise signals can significantly interfere with the extraction of characteristic signals, reduce the signal-to-noise ratio of the received echo signal, introduce errors into the monitoring results, and have a significant adverse impact on the monitoring of equipment defects.

[0003] Generally, hardware denoising alone cannot completely and effectively eliminate noise. Instead, denoising and purification based on the characteristics of the acquired signal using appropriate algorithms is a more practical and efficient method. The paper "Research on Denoising of Ultrasonic Monitoring Signals Based on Variational Mode Decomposition" (China Testing, 2019, 45(12):106-111) proposes a denoising method for ultrasonic monitoring signals. This method achieves denoising of ultrasonic monitoring signals in four steps: variational mode decomposition, calculation of the bandwidth and center frequency of each mode, selection of signal modes, and reconstruction of a noiseless signal based on the signal modes. It has a good effect on denoising complex ultrasonic signals containing different mode bandwidths and center frequencies. CN108491355A discloses an ultrasonic signal denoising method based on CEEMD and wavelet packets. This method alleviates the problem of mode aliasing and makes the decomposition of multi-mode signals more thorough.

[0004] However, these noise reduction techniques mainly address the multimodal and aliasing problems of ultrasonic signals, but are ineffective at reducing environmental noise in engineering applications. To solve this problem, this invention provides a signal purification method for an online waveguide monitoring system, which can effectively reduce noise in ultrasonic signals acquired in practical engineering projects. Summary of the Invention

[0005] The technical problem this invention aims to solve is the difficulty of measurement error caused by the susceptibility of ultrasonic signals to ambient noise in harsh industrial environments with numerous interference factors, leading to impure monitoring signals. This invention provides a method for purifying ultrasonic signals in high-temperature industrial environments based on wavelet denoising. When performing long-term monitoring of pressure equipment, the method performs noise reduction processing on the 5-cycle sinusoidal ultrasonic signal modulated by the Hanning window acquired by the waveguide rod online monitoring system, thereby improving measurement accuracy.

[0006] This invention is achieved through the following technical solution:

[0007] Ultrasonic testing is an active excitation type of testing, and the acquired echo signal still contains information from the original excitation signal. A signal purification method is provided for the 5-cycle sinusoidal pulse signal modulated by the Hanning window used in the ultrasonic guide rod monitoring system. The purification process mainly includes the following steps:

[0008] Step 1: Perform wavelet decomposition on the acquired ultrasonic signal. Considering the abrupt and non-stationary characteristics of ultrasonic pulse signals, the time-domain compact support should be fully considered when selecting the wavelet basis function. At the same time, a fast decay rate is required. Based on the self-similarity principle, after multiple tests and comparisons, the db8 wavelet was selected as the wavelet basis function for discrete wavelet transform.

[0009] Step 2: After multiple layer comparisons, when the signal was decomposed into 7 layers, it was found that the low-frequency signal still retained the original 5-cycle sine wave signal, and the noise signal mixed in the low-frequency part was also separated out. Therefore, 7 layers were considered to be a better decomposition layer.

[0010] Step 3: After decomposing the signal into 7 levels, the high-frequency components are quite cluttered. Therefore, a Stein-based unbiased risk estimation method is used to determine the thresholds for the high-frequency coefficients of each level to reduce noise.

[0011] Step 4: Use an exhaustive method to readjust the threshold. The new threshold is the result of optimizing the product of the threshold determined based on the unbiased risk estimation and the variance of the first-layer high-frequency noise.

[0012] Step 5: Using the soft thresholding method, the absolute value of each decomposition layer coefficient is compared with the threshold. If the absolute value is smaller than the threshold, the coefficient is set to 0. If the absolute value is larger than the threshold, the value is subtracted from the threshold to obtain a new coefficient.

[0013] Step Six: Reconstruct and synthesize the signals from each layer that have been reprocessed by selecting the combination of "rigrsure, sln, S" to achieve the effect of noise reduction and purification.

[0014] Beneficial effects

[0015] The advantages of this invention are:

[0016] This invention can quickly and effectively purify and reduce noise in a 5-cycle sinusoidal ultrasonic signal with noisy, non-stationary Hanning window modulation.

[0017] This invention has good time-frequency characteristics and does not change the key index characteristics of non-destructive monitoring such as the phase and shape of the original signal, thus ensuring the accuracy of measurement during corrosion thinning test.

[0018] This invention can effectively reduce noise even under high-temperature conditions. In an environment with a temperature gradient of 50 to 350°C, the signal-to-noise ratio of ultrasonic signals is greatly improved, and the original characteristic signals are also preserved, thereby improving the monitoring accuracy. Attached Figure Description

[0019] Figure 1 This is a flowchart of the signal purification framework of the present invention.

[0020] Figure 2 Five-cycle sinusoidal pulse ultrasonic signals modulated by the Hanning window in the waveguide rod online monitoring system

[0021] Figure 3(a) shows the db8 scaling function graph.

[0022] Figure 3(b) shows the db8 wavelet function graph.

[0023] Figure 4 Ultrasonic signal graph with 3% added noise

[0024] Figure 5 To filter the graph using db8 wavelet 7-level wavelet decomposition

[0025] Figure 6 A diagram of the soft thresholding method.

[0026] Figure 7 A comparison chart of a signal before and after noise reduction.

[0027] Figure 8 Implementation Case System Framework Diagram

[0028] Figure 9(a) shows the comparison of signal noise reduction before and after at 50℃.

[0029] Figure 9(b) shows the comparison of signal noise reduction before and after at 100℃.

[0030] Figure 9(c) shows the comparison of signal noise reduction before and after at 150℃.

[0031] Figure 9(d) shows the comparison of signal noise reduction before and after at 200℃.

[0032] Figure 9(e) shows the comparison of signal noise reduction before and after at 250℃.

[0033] Figure 9(f) shows the comparison of signal noise reduction before and after at 300℃.

[0034] Figure 9(g) shows the comparison of signal noise reduction before and after at 350℃.

[0035] Figure 10 A schematic diagram of the signal-to-noise ratio calculation method.

[0036] Figure 11 Comparison of signal-to-noise ratio before and after denoising at different ambient temperatures Detailed Implementation

[0037] The ultrasonic signal purification and analysis method of the present invention will be further described in detail below with reference to the accompanying drawings.

[0038] The waveguide rod online monitoring system uses an ultrasonic signal, which is an active excitation type of monitoring. The acquired raw echo signal contains both initial excitation signal and noise signal information. The raw excitation signal used in this invention is a 5-cycle sinusoidal pulse signal modulated by a Hanning window. This type of windowed signal has advantages such as reducing spectral leakage, suppressing dispersion after sampling, and reducing high-frequency signal interference to a certain extent. See [link to raw excitation signal] for details. Figure 2 The mathematical expression for this signal is:

[0039]

[0040] In the formula, f is the frequency of the excitation signal, in Hertz (Hz); t is the time, in seconds (s); and n is the number of signal cycles, in this application n = 5.

[0041] Considering the abrupt change of ultrasonic pulse signals, the time-domain compact support should be fully considered when selecting wavelet basis functions, and a fast attenuation rate is also required. Based on the principle of self-similarity, after multiple tests and comparisons of filtering effects, it was found that the db8 wavelet is the best choice as the basis function. The scaling function and wavelet function of the db8 wavelet are shown in Figure 3.

[0042] To optimize the maximum number of decomposition layers and the selection threshold, a set of sinusoidal ultrasonic signals modulated by a Hanning window with good signal-to-noise ratio were selected, and 3% white noise was added. The signal after adding white noise can be found here. Figure 4 In the image, only the signals of the first two wave packets can be seen with the naked eye; the signals of subsequent wave packets are completely annihilated and cannot be identified.

[0043] The noisy signal was decomposed using the db8 wavelet with a scale of 7 using discrete wavelet decomposition. The wavelet decomposition selection chart is shown below. Figure 5 At this point, the original signal s is decomposed into signals at scales a7 and d1 to d7. The low-frequency scale signal a7 in the figure is very similar to the original 5-cycle sine wave signal, and the noise mixed in with the original signal is also separated out, while the high-frequency scale signals d1 to d7 are more chaotic. If the number of decomposition layers is too large, feature signals will be lost, signal strength will decrease significantly, and computational redundancy will slow down the processing speed. If the number of decomposition layers is too small, noise cannot be separated from the target signal, resulting in poor noise reduction.

[0044] For the high-frequency coefficients of each layer after discrete wavelet decomposition, take the absolute value of each element, sort them in ascending order, and then square them. This will give a new sequence, as shown in the following formula:

[0045] f(k) = (sort(|s|)) 2(k=0,1,2...,N-1) (2)

[0046] If the threshold is defined as the square root of an element in f(k), then we have The risk at this threshold is:

[0047]

[0048] From the risk curve Risk(k) obtained by the above formula, we can find the element k corresponding to the lowest risk point. min Then the threshold for Stein's unbiased risk estimation formula is...

[0049] The calculated threshold λ is multiplied by the variance of the high-frequency scale signal noise in the first layer of decomposition, and the result is used as the new readjustment threshold λ.

[0050] The wavelet coefficients of each layer are processed using the soft thresholding method, and the expression used is:

[0051]

[0052] If the absolute value of the coefficient is smaller than the readjustment threshold λ, the coefficient is directly set to 0. If it is larger than the threshold, the value is subtracted from the threshold to obtain a new coefficient. See the schematic diagram for the process. Figure 6 .

[0053] The reprocessed coefficients of each layer of signals are reconstructed into a composite signal to achieve noise reduction and purification. This method was used to perform noise reduction and purification on the aforementioned feature signal containing 3% noise. A comparison of the actual noise reduction results is shown below. Figure 7 As can be seen, at least four waveforms can be clearly identified in the image after noise reduction. This noise reduction effect has great practical significance in some cases where the return signal is weak and in ultrasound imaging.

[0054] Implementation Case:

[0055] To verify the reliability of the signal purification method for an online waveguide rod monitoring system proposed in this invention, an online waveguide rod monitoring test system was created as follows: Figure 8 As shown, the system consists of a commercial signal generator, power amplifier, digital oscilloscope, waveguide-type piezoelectric sensor, 316L stainless steel plate (400mm×300mm×1mm), and high-temperature chamber. The 316L stainless steel plate is placed inside the high-temperature chamber as the test piece. The proximal end face of the waveguide rod of the waveguide-type piezoelectric sensor is fixed to the test piece. The waveguide rod passes through a hole on the upper surface of the high-temperature chamber and extends out of the chamber, while the distal end face of the piezoelectric sensor is extended out of the chamber. The function of the waveguide rod is to isolate the high-temperature environment from the piezoelectric sensor, protecting the sensor from the effects of high temperature.

[0056] The signal generator in the experiment was used to excite a 5-cycle sinusoidal ultrasonic signal modulated by a Hanning window. A power amplifier amplified the excitation signal output from the signal generator, and a digital oscilloscope was used to acquire the ultrasonic echo signal. Two waveguide-type piezoelectric sensors were used in the experimental system, one for excitation and one for reception. The noise reduction method of this invention was used to process the ultrasonic signals acquired by the waveguide-type online monitoring system under different temperature environments. Data was collected at 25°C intervals from 50°C to 350°C, for a total of 13 sets. Data at temperatures of 50°C, 100°C, 150°C, 200°C, 250°C, 300°C, and 350°C were extracted and plotted in Figure 9. Figures 9(a)-9(g) It can be seen that after wavelet denoising, the ultrasonic signals propagating in the waveguide rod online monitoring system at different temperatures are purified, while the characteristic signals are still preserved and the time domain signals before and after remain unchanged. Clutter at a distance from the characteristic signals has been basically suppressed.

[0057] The ultrasonic signal in Figure 9 consists of two wave packets. The first signal wave packet is the direct wave between the two waveguide-type piezoelectric sensors, which is a useful signal in the non-destructive testing process. The second signal wave packet is a composite wave packet formed by the reflected echoes from the left and right edges. This wave packet contains more complex information and is not considered in this test. Based on the original echo signal at 50℃, using the formula... Calculate the signal-to-noise ratio (SNR) of the signal. A0 represents the maximum amplitude of the echo signal, and the area within the dashed box after the echo represents clutter. Divide this segment of the signal into several parts based on its noise characteristics, and average the maximum amplitude of each part to obtain A1. See the schematic diagram below. Figure 10 ,

[0058] Use formula The signal-to-noise ratio of the ultrasonic signal before and after noise reduction at different temperatures was calculated. See the results below. Figure 11 As can be seen, the noise reduction method of this invention significantly improves the signal-to-noise ratio (SNR) of ultrasonic signals at different temperatures. The optimal SNR of the processed signal is close to 25 dB, and the minimum is above 16 dB, meeting the requirements of practical engineering. Furthermore, comparison of waveforms and data before and after noise reduction reveals that this method does not alter the key characteristics of the original signal, such as phase and morphology, in online monitoring. It has high practical value for fields such as ultrasonic thickness measurement, flaw detection, and imaging.

Claims

1. A method for signal purification in an online monitoring system for waveguide rods, characterized in that, The five-cycle sinusoidal pulse ultrasonic signal modulated by the Hanning window, acquired by the waveguide rod online monitoring system, is decomposed into seven layers using the db8 wavelet, and the wavelet coefficients of each layer are thresholded to reconstruct the signal. The process includes the following steps: Step 1: Perform wavelet decomposition on the acquired 5-cycle sinusoidal ultrasonic signal with Hanning window modulation containing noise, and select the db8 wavelet as the wavelet basis function for discrete wavelet transform. Step 2: Perform a 7-layer decomposition on the 5-cycle sinusoidal ultrasonic signal modulated by the Hanning window to obtain 1 layer of low-frequency coefficients and 7 layers of high-frequency coefficients; Step 3: Determine the threshold for each high-frequency coefficient obtained from the decomposition using an unbiased risk estimation method based on Stein. Specifically, for each high-frequency coefficient, take the absolute value of each element, sort them in ascending order, and then square them to obtain a new sequence, as shown in the following formula: （1）， Define the threshold as the square root of an element in f(k), and we have a preliminary threshold. At this point, the risk of this threshold is: （2）， From the risk curve Risk(k) obtained by the above formula, we can find the element k corresponding to the lowest risk point. min Then, the threshold for unbiased risk estimation based on Stein is... ; Step 4: Readjust the threshold New threshold The product of the threshold determined for unbiased risk estimation and the variance of the first-level high-frequency noise; Step 5: Using a soft thresholding method, compare the absolute value of each decomposition layer coefficient with the threshold. If the value is smaller than the threshold, set the coefficient to 0; if the value is larger than the threshold, subtract the threshold to obtain the new coefficient. As shown in the following formula: （3）； Step 6: Reconstruct the ultrasonic signal based on the low-frequency coefficients of wavelet decomposition and the 7 layers of high-frequency coefficients after thresholding to obtain the noise-reduced signal.

Images