A method for predicting valve internal leakage level based on acoustic emission signals

Abstract

The present invention discloses a method for predicting the internal leakage level of a valve based on an acoustic emission signal, which solves the problems in the prior art of being unable to automatically extract valve internal leakage features, as well as the problems of low modeling efficiency and low prediction accuracy. The method comprises: first, collecting the acoustic emission signals of the valve under different internal leakage rates, and performing bandpass filtering preprocessing on the original signals. Then, wavelet scattering transform is used to extract features of the valve internal leakage acoustic emission signal, the second-order scattering coefficient is converted into a two-dimensional feature matrix by averaging in the time dimension, and the ReliefF algorithm is used to extract the optimal scattering coefficient feature. The optimal feature subset and pressure are used as inputs of the classification model, and the AdaBoost.M1 method is used for classification modeling. Finally, the trained AdaBoost.M1 model is used to predict the internal leakage level of the unknown valve internal leakage acoustic emission signal. The present invention realizes the automatic extraction of valve internal leakage features and improves the efficiency and prediction accuracy of valve internal leakage acoustic emission signal modeling.

Description

Technical Field

[0001] The present invention relates to the fields of valve internal leakage diagnosis technology and signal processing and analysis technology, and in particular to a valve internal leakage level prediction method based on acoustic emission signals. Background Art

[0002] Valves are key control components in pipeline operations and are widely used in industries such as oil, natural gas, nuclear power, and water conservancy. Because valves often operate under high-temperature and high-pressure conditions and are subject to impact and wear from the media in the pipeline, they are prone to internal leakage.

[0003] In order to reduce the economic losses and industrial safety accidents caused by valve internal leakage, domestic and foreign scholars have studied and established a variety of valve internal leakage diagnosis methods. Among them, acoustic emission technology has become a widely studied valve internal leakage detection method because it can be detected online without removing the valve from the pipeline, and the detection process is convenient and efficient. The paper “Deep belief network-based internal valve leakage rate prediction approach” (Zhu SB, Measurement, 2019, Vol. 133) collected valve data of two types of valves, plug valves and ball valves, and extracted 16 time domain features and 5 frequency domain features of the acoustic emission signal as the input of the model; the paper “Multivariable modeling of valve inner leakage acoustic emission signal based on Gaussian process” (Ye GY, Mechanical Systems and Signal Processing, 2020, Vol. 140) extracted the standard deviation, spectral area, wavelet packet entropy and other features of the acoustic emission signal as features in order to establish a multivariate regression model of the acoustic emission signal of small leakage of liquid medium valves; the paper “Quantification of valve leakage rates” (Meland E, AIChE journal, 2012, Vol. 58) selected the root mean square, spectral component and independent component analysis mixed vector of the signal as features in order to quantify the internal leakage rate of the stop valve; the paper “Application of Acoustic Emission and Support Vector Machine to Detect the Leakage of Pipeline Valve” (Zhang HF, 2013 Fifth International Conference on Measuring Technology and Mechatronics Automation, Hong Kong, China, 2013) used the root mean square (RMS) of the acoustic emission signal, the average signal level in the time domain, and the peak value in the frequency domain as inputs to the support vector machine model. All of these studies manually selected some time- and frequency-domain features of the AE signal. While some of these features have been shown to have theoretical relationships with leakage rates, many were selected based on professional experience, which can lead to unsatisfactory modeling results. Furthermore, manually extracting and filtering signal features is time-consuming and reduces modeling efficiency.

[0004] The paper “A novel acoustic emission detection module for leakage recognition in a gas pipeline valve” (Li ZL, Process Safety and Environmental Protection, 2017, Vol. 105) proposed a leakage detection method based on kernel principal component analysis and support vector machine, used kernel principal component analysis to reduce the dimension of features, and used support vector machine to identify 8 leakage levels; the paper “Multi-variable classification model for valve internal leakage based on acoustic emission time–frequency domain characteristics and random forest” (YeG Y, Review of Scientific Instruments, 2021, Vol. 92) aimed at the problem that the acoustic emission signal of liquid medium is greatly interfered by noise, used Butterworth bandpass filter for noise reduction, and used random forest method to establish a multivariate classification model between acoustic emission signal characteristics and leakage rate under variable pressure to predict the leakage level of the valve; the paper “Detection and estimation of valve leakage losses in reciprocating compressorusing acoustic emission technique” (Sim HY, Measurement, 2020, Vol. 152) et al. used k-nearest neighbor and support vector machines to classify valve states and found that support vector machines achieved higher classification accuracy. Existing valve internal leakage acoustic emission signals are modeled using a single learner, resulting in low model accuracy and poor robustness.

[0005] In summary, existing technologies all rely on manual extraction of valve internal leakage features, resulting in long feature extraction cycles and poor noise immunity. Furthermore, a single learner is used to model the valve internal leakage acoustic emission signal, resulting in low robustness and low accuracy in predicting valve internal leakage levels. Therefore, automated extraction of valve internal leakage acoustic emission signal features and improved model accuracy in leak level prediction are urgent challenges. Summary of the Invention

[0006] The purpose of the present invention is to solve the problems that the features of the existing valve internal leakage acoustic emission signal need to be manually extracted and the extraction cycle is long, and the use of a single learner for modeling leads to poor model robustness and low prediction accuracy. A valve internal leakage level prediction method combining wavelet scattering transform and AdaBoost.M1 is proposed, thereby improving the modeling speed and prediction accuracy of the acoustic emission signal under complex background noise.

[0007] The present invention is achieved through the following technical solutions:

[0008] Step 1: Using a valve internal leakage detection device to sample the acoustic emission signal at the valve body under different valve internal leakage rate conditions, the data includes the acoustic emission signal data of different degrees of small angle opening when the valve simulates internal leakage and the corresponding prior internal leakage rate information, and generate a data set;

[0009] Step 2: The acoustic emission signal collected in step 1 is pre-processed by using a Butterworth bandpass filter to reduce noise;

[0010] Step 3: The pre-processed signal is subjected to a wavelet scattering transform to obtain the third-order scattering coefficients S0, S1, and S2. The scattering coefficients are converted into a one-dimensional feature vector by averaging the scattering window dimensions.

[0011] Step 4: Select appropriate features through the ReliefF algorithm to reduce the redundancy of high-dimensional features, and the remaining features form a feature vector data set;

[0012] Step 5: Normalize the data set obtained in step 4 and process it using the "k-fold cross validation" method to generate a combination of training set and validation set;

[0013] Step 6: Use the training set sample data in step 5 to train the AdaBoost.M1 model, and use the validation set for verification, and finally determine the number of AdaBoost.M1 iterations;

[0014] Step 7: The features of the valve internal leakage acoustic emission signal with unknown internal leakage level to be tested are extracted using the method of steps 2 to 4, and the features are input into the AdaBoost.M1 model with a determined structure generated in step 6 to predict the valve internal leakage level, thereby realizing the valve internal leakage level prediction based on acoustic emission technology.

[0015] Preferably, the valve internal leakage detection device consists of an acoustic emission sensor, a voltage amplifier, an anti-aliasing filter and a data acquisition card. The acoustic emission sensor is connected to the voltage amplifier, the voltage amplifier is connected to the anti-aliasing filter, the anti-aliasing filter is connected to the data acquisition card, and the data acquisition card is connected to the computer; the acoustic emission sensor is installed on the middle platform of the valve body of the valve to be tested, and the acoustic emission sensor is also fixed by a magnetic clamp, and a coupling agent is applied between the acoustic emission sensor and the valve body.

[0016] Preferably, the frequency range of the acoustic emission sensor is 50-400kHz; the voltage amplifier is a PAI type preamplifier with a gain of 40dB and an output voltage range of ±10V; the cutoff frequency of the anti-aliasing filter is 600kHz; and the maximum sampling frequency of the data acquisition card is 1.25MHz.

[0017] Preferably, step 1 specifically refers to: under different pressure environments, simulating n different small valve opening states with different internal leakage rates, respectively collecting the acoustic emission signal at the middle platform of the current valve body and the prior internal leakage rate data to generate a sampling data set X:

[0018] X={x1, x2, …, x n} (1)

[0019] Preferably, the step 2 specifically refers to: after fast Fourier transform (FFT) analysis, the sampled acoustic emission signal is bandpass filtered to obtain a preprocessed data set X r :

[0020] X r ={x r1 , x r2 , …, x rn} (2)

[0021] Preferably, step 3 specifically includes: first setting a wavelet scattering network structure, setting the quality factor of the first layer of the scattering network to 2, the quality factor of the second layer to 1, and the time invariance scale to 0.1S. Then, combining the obtained 0th-order, 1st-order, and 2nd-order scattering coefficients and averaging them in the time dimension to form a scattering coefficient feature dataset.

[0022] Preferably, the step 4 specifically refers to: using the ReliefF algorithm to sort the features according to their weights, and retaining the 10 features with the largest weights, and adding pressure to form the optimal feature data set D m .

[0023] D m ={d1, d2, ..., d m} m≤n (3)

[0024] Preferably, the step 5 specifically refers to: the normalization formula is:

[0025]

[0026] Where D m (i) is the extracted feature, D norm is the normalized feature, D max 、Dmin are the maximum and minimum values ​​in the data series, respectively.

[0027] "5-fold cross validation" method: Divide the dataset into 5 mutually exclusive subsets of equal size, use the first 4 subsets as training sets each time, and the remaining subsets as test sets.

[0028] Preferably, step 6 specifically comprises: using a "5-fold cross validation" method to train the model for 500 iterations. When the accuracy of the validation set is relatively the highest, the number of iterations of the current AdaBoost.M1 model is determined.

[0029] Preferably, step 7 specifically involves using the trained AdaBoost.M1 model to predict the valve internal leakage level of the test data. Preprocessing the acoustic emission signal of the unknown internal leakage rate to be measured, extracting eigenvalues ​​using wavelet scattering transform, and normalizing them to construct a standard eigenvector. This eigenvector is then input into the AdaBoost.M1 model with a defined structure to predict the valve internal leakage level, thereby achieving valve internal leakage level prediction based on the acoustic emission signal.

[0030] Compared with the prior art, the advantages of the present invention are:

[0031] (1) In the present invention, the wavelet scattering transform is combined with the ReliefF algorithm for feature extraction. The features extracted by the wavelet scattering transform have translation invariance and local deformation stability, and have good physical interpretation capabilities. The feature extraction process does not require manual intervention, which overcomes the problem of relying on professional experience and time-consuming manual extraction of time-frequency domain features in previous models, increases the efficiency of feature extraction, and speeds up modeling.

[0032] (2) In this invention, an ensemble learning method is used for modeling. Compared with a single learner, the AdaBoost.M1 method has higher accuracy and stronger generalization ability. This improves the accuracy and stability of the valve internal leakage level prediction model. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is the overall architecture diagram of the valve internal leakage level prediction method based on acoustic emission technology proposed in the present invention;

[0034] Figure 2 Schematic diagram of the valve internal leakage detection experimental platform based on acoustic emission sensor in the present invention.

[0035] Figure 3 Schematic diagram of wavelet scattering transform decomposition in the present invention.

[0036] Figure 4 This is the feature weight ranking diagram calculated by the ReliefF algorithm in the present invention.

[0037] Figure 5 Schematic diagram of the iteration of the AdaBoost.M1 model in the present invention. DETAILED DESCRIPTION

[0038] The present invention is further described in detail below with reference to the accompanying drawings:

[0039] Reference Figure 1 , Figure 1 A flow chart of a method for predicting valve internal leakage level based on acoustic emission signals is drawn. As shown in the figure, a method for predicting valve internal leakage level based on acoustic emission signals of the present invention includes the following steps:

[0040] (1) Data acquisition, valve internal leakage detection experimental platform based on acoustic emission sensor, such as Figure 2 As shown. The acoustic emission signals at the valve body under different valve internal leakage rate conditions are sampled. The data includes the acoustic emission signal data of n different degrees of small angle opening when the valve simulates internal leakage and the corresponding prior internal leakage rate information, generating a sampling data set X:

[0041] X={x1, x2, …, x n} (1)

[0042] (2) Data preprocessing: The time domain signal collected in step 1 is preprocessed by a 40kHz-320kHz bandpass filter. The filter used is a fourth-order Butterworth bandpass filter to obtain the preprocessed data set X r :

[0043] X r ={x r1 , x r2 , …, x rn} (2)

[0044] (3) Feature extraction: Perform wavelet scattering transform on the pre-processed signal to obtain the 0th, 1st and 2nd order scattering coefficients. The wavelet scattering transform flow chart is as follows: Figure 3 As shown. The calculation formula for the n-order scattering coefficient is:

[0045]

[0046] Where x is the input signal, ψ is the set of mother wavelets, whose center frequency is λ, represents the convolution operation, is the scaling function;

[0047] The scattering coefficient is averaged in the time dimension to form a scattering coefficient feature data set D nThe quality factor of the first layer of the scattering network is set to 2, the quality factor of the second layer is set to 1, and the time invariance scale is set to 0.1s.

[0048] (4) Feature selection, the feature dataset D in step 3 n The ReliefF method is used to screen and select, preprocess the original feature set, remove redundant information, increase the representation ability of the remaining vectors, and form a new feature data set D m The feature weight is as follows: Figure 4 shown.

[0049] D m ={d1, d2, ..., d m} m≤n (4)

[0050] (5) Feature normalization: Considering the difference in data dimensions between input and output data, the feature dataset D m Normalize it with the prior valve internal leakage rate information (label) to form a new data set D norm , for the dataset D norm The "k-fold cross validation" method is used to generate k sets of training and test sets. Normalization generally transforms the data to between [0,1], where the normalization formula is:

[0051]

[0052] Where D m (i) is the extracted feature, D norm is the normalized feature, D max 、D min are the maximum and minimum values ​​in the data series, respectively.

[0053] (6) Model training: All samples obtained in step 5 are used to train the AdaBoost.M1 model. During training, the data set is divided into 5 mutually exclusive subsets of equal size. The first 4 subsets are used as training sets each time, and the remaining subsets are used as test sets. The training is repeated 500 times. When the accuracy of the validation set is relatively high, the number of iterations L of the current AdaBoost.M1 model is determined. The training process of the AdaBoost.M1 model is as follows: Figure 5 shown.

[0054] (7) Leakage level prediction: Use the trained AdaBoost.M1 model to detect the internal leakage level of the valve on the data to be tested. The acoustic emission signal of the unknown internal leakage rate to be tested is preprocessed, and the eigenvalues ​​are extracted using wavelet scattering transform. The standard eigenvector is constructed by normalization and input into the AdaBoost.M1 model with a certain structure. The current valve internal leakage level is predicted to be small, medium or large, thereby realizing the valve internal leakage level detection based on the acoustic emission signal.

[0055] The above are only specific application examples of the present invention and do not constitute any limitation on the scope of protection of the present invention. Any technical solutions formed by equivalent transformation or equivalent replacement shall fall within the scope of protection of the present invention.

Claims

1. A method for predicting valve internal leakage level based on acoustic emission signals, characterized in that: The steps include: Step 1: Using a valve internal leakage detection device to sample the acoustic emission signal at the valve body under different valve internal leakage rate conditions, the data includes the acoustic emission signal data of different degrees of small angle opening when the valve simulates internal leakage and the corresponding prior internal leakage rate information, and generate a data set; Step 2: The acoustic emission signal collected in step 1 is pre-processed by using a Butterworth bandpass filter to reduce noise; Step 3: The preprocessed signal is subjected to wavelet scattering transform to obtain the third-order scattering coefficient; the scattering coefficient is converted into a one-dimensional feature vector by averaging the scattering window dimension; appropriate features are selected using the ReliefF algorithm to reduce the redundancy of high-dimensional features, and the remaining features form the feature vector data set; Step 4: Normalize the dataset obtained in step 3 and process it using the "k-fold cross-validation" method to generate a combination of training and validation sets; the training set sample data is used to train the AdaBoost.M1 model, and the validation set is used for validation, ultimately determining the number of AdaBoost.M1 iterations; Step 5: The features of the valve internal leakage acoustic emission signal with unknown internal leakage level to be tested are extracted using the method of steps 2 to 4, and input into the generated AdaBoost.M1 model with a fixed structure to predict the valve internal leakage level, thereby realizing the valve internal leakage level prediction based on acoustic emission technology.

2. The method for predicting valve internal leakage level based on acoustic emission signals according to claim 1, characterized in that: The valve internal leakage detection device consists of an acoustic emission sensor (1), a voltage amplifier (2), an anti-aliasing filter (3) and a data acquisition card (4). The acoustic emission sensor (1) is connected to the voltage amplifier (2), the voltage amplifier (2) is connected to the anti-aliasing filter (3), the anti-aliasing filter (3) is connected to the data acquisition card (4), and the data acquisition card (4) is connected to a computer. The acoustic emission sensor (1) is installed on the middle platform of the valve body of the valve to be tested. The acoustic emission sensor (1) is also fixed by a magnetic clamp, and a coupling agent is applied between the acoustic emission sensor and the valve body.

3. The method for predicting valve internal leakage level based on acoustic emission signals according to claim 1, characterized in that: The step 3 specifically includes: first setting the wavelet scattering network structure, setting the quality factor of the first layer of the scattering network to 2, the quality factor of the second layer to 1, and the time invariance scale to 0.1S; and combining the obtained 0th order, 1st order, and 2nd order scattering coefficients and averaging them in the time dimension to form a scattering coefficient feature data set D n ; Use the ReliefF algorithm to sort the features according to their weights, and retain the 10 features with the largest weights, plus the pressure, to form the optimal feature dataset D m ; D m ={d1 ,d2 ,...,d m } m≤n (1) The normalization formula is: Where D m (i) is the extracted feature, D norm is the normalized feature, D max 、D min are the maximum and minimum values ​​in the data series, respectively.

4. The method for predicting valve internal leakage level based on acoustic emission signals according to claim 1, characterized in that: The step 5 specifically includes: using the trained AdaBoost.M1 model to predict the valve internal leakage level of the test data; preprocessing the acoustic emission signal of the unknown internal leakage rate to be measured, extracting eigenvalues ​​using wavelet scattering transform, and normalizing them to construct a standard eigenvector, which is input into the AdaBoost.M1 model with a determined structure to predict the valve internal leakage level, thereby realizing the valve internal leakage level prediction based on the acoustic emission signal.

Images