Pipeline leakage diagnosis method based on sound signal and vibration signal fusion network

Abstract

The invention relates to the technical field of pipeline leakage, in particular to a pipeline leakage diagnosis method based on an acoustic signal and vibration signal fusion network, which comprises the following steps: constructing a fusion classifier comprising an acoustic emission classifier, a vibration classifier and a combined classifier; feature extraction networks of the acoustic emission classifier and the vibration classifier are a dense connection layer, a first activation layer and a regularization layer of a stacked structure; a feature extraction network of the combined classifier is formed by cascading a compression attention mechanism module with a dense connection layer and a first activation layer of a stacked structure; a classification network of the combined classifier is formed by cascading a dense connection layer and a second activation layer; the sound signal feature vector, the vibration feature vector and the knocking angle value serve as input of a combined classifier; and performing residual connection on the feature extraction network of the combined classifier and the dense connection layer of the classification network thereof. The problem that in the prior art, the pipeline leakage judgment accuracy of a single signal needs to be further improved is solved.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline leakage, and in particular to a pipeline leakage diagnosis method based on an acoustic signal and a vibration signal fusion network. Background Art

[0002] In view of the potential dangers of pipeline gas leakage, timely and accurate diagnosis is particularly important; through effective diagnostic methods, gas leakage points can be discovered in time and appropriate remedial measures can be taken to avoid the occurrence of safety accidents.

[0003] Patent application number 202510021350.4 is a gas pipeline leakage diagnosis method based on imbalance data. It only uses acoustic emission signals. Although acoustic emission signals can capture high-frequency transient events, they do not pay enough attention to low-frequency mechanical vibrations that can reflect structural abnormalities, looseness and other factors. Vibration signals can reflect the overall operating status of the mechanical structure under low-frequency conditions, but for gas pipeline leakage processes with different leakage severities, due to their weak characteristics, they are difficult to identify and have low accuracy. Summary of the invention

[0004] In view of the shortcomings of the existing methods, the present invention constructs a fusion classifier composed of an acoustic emission classifier, a vibration classifier and a combined classifier, and fuses the acoustic signal feature vector, the vibration feature vector and the knocking angle to improve the accuracy of pipeline leakage classification.

[0005] The technical solution adopted by the present invention is: a pipeline leakage diagnosis method based on a fusion network of acoustic signals and vibration signals comprises the following steps:

[0006] Step 1: Collect acoustic emission signals and vibration signals at different knocking angles to construct a pipeline leakage status data set;

[0007] As a preferred embodiment of the present invention, different leakage states are simulated by adjusting the horizontal angle between the percussion hammer and the pipeline.

[0008] As a preferred embodiment of the present invention, the leakage status includes: healthy, slight leakage, moderate leakage and severe leakage.

[0009] Step 2: preprocessing the pipeline leakage status data set;

[0010] As a preferred embodiment of the present invention, the preprocessing includes: calculating the mean value, the standard deviation value, standardization and fast Fourier transform.

[0011] Step three, construct a fusion classifier including an acoustic emission classifier, a vibration classifier and a combined classifier; the feature extraction networks of the acoustic emission classifier and the vibration classifier are densely connected layers, first activation layers and regularization layers of a stacked structure; the feature extraction network of the combined classifier is a cascade of a compressed attention mechanism module with a densely connected layer and a first activation layer of a stacked structure; the classification network of the combined classifier is a cascade of a densely connected layer and a second activation layer; the acoustic signal feature vector, the vibration feature vector and the knocking angle value are used as inputs of the combined classifier; the feature extraction network of the combined classifier is residually connected to the densely connected layer of its classification network.

[0012] As a preferred embodiment of the present invention, the stacking structure of the feature extraction network of the acoustic emission classifier is 3 layers.

[0013] As a preferred embodiment of the present invention, the stacking structure of the feature extraction network of the vibration classifier is 2 layers.

[0014] As a preferred embodiment of the present invention, the stacking structure of the feature extraction network of the combined classifier is 2 layers.

[0015] As a preferred implementation of the present invention, the first activation layer is a ReakyRelu activation function, and the second activation layer is a SoftMax activation function.

[0016] As a preferred embodiment of the present invention, a pipeline leakage diagnosis system based on a fusion network of acoustic signals and vibration signals includes: a memory for storing instructions executable by a processor; and a processor for executing the instructions to implement a pipeline leakage diagnosis method based on a fusion network of acoustic signals and vibration signals.

[0017] As a preferred embodiment of the present invention, a computer readable medium stores a computer program code, and when the computer program code is executed by a processor, a pipeline leakage diagnosis method based on a fusion network of acoustic signals and vibration signals is implemented.

[0018] Beneficial effects of the present invention:

[0019] 1. Build a fusion classifier to fuse the acoustic signal feature vector, vibration feature vector and tapping angle, and use the compressed attention mechanism module to capture the correlation of the three feature vectors. Perform residual design on the densely connected layer of the combined classifier to make the fusion classifier efficiently trained, converge quickly, and improve generalization ability. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a structural diagram of the fusion classifier of the present invention;

[0021] Figure 2 is a schematic diagram of the striking angle of the striking hammer of the present invention;

[0022] Figure 3 It is a schematic diagram of the sensor and the percussion hammer of the present invention;

[0023] Figure 4 is a confusion matrix diagram of the present invention;

[0024] Figure 5 It is a comparison diagram of the individual model, the fusion model and the existing support vector machine of the present invention. DETAILED DESCRIPTION

[0025] The present invention is further described below in conjunction with the accompanying drawings and embodiments. This figure is a simplified schematic diagram, which only illustrates the basic structure of the present invention in a schematic manner, and therefore it only shows the components related to the present invention.

[0026] like Figure 1 As shown, the pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network includes the following steps:

[0027] Step 1: Perform a tapping test on the surface of the gas pipeline to simulate the operating conditions; collect acoustic emission signals and vibration signals at different tapping angles to construct a data set;

[0028] Use a percussion hammer to perform a percussion test on a specified position on the surface of the gas pipeline, control the horizontal angle between the percussion hammer and the surface of the gas pipeline, and simulate the percussion intensity;

[0029] like Figure 2 , 3 , a percussion test is carried out on the gas pipeline to simulate different operating conditions; the leakage status is divided into four types: healthy, slight leakage, moderate leakage and severe leakage; when the percussion hammer is placed on the surface of the gas pipeline with a horizontal angle of 0°, the pipeline is in a healthy state; when the angle is 30° and a percussion is applied, a slight leakage is simulated; when the angle increases to 60°, it corresponds to a moderate leakage; if the angle reaches 90°, it represents a severe leakage state.

[0030] A combined system using the low-frequency vibration signal of an accelerometer and the high-frequency acoustic emission signal of an ultrasonic transducer, that is, using an acceleration sensor to measure the vibration signal of the equipment, and using an acoustic emission sensor to measure the acoustic emission signal of the equipment. The types and quantity of sensors may not be limited to these two types.

[0031] Step 2: Preprocess the data set;

[0032] Preprocess the data set, calculate the mean and standard deviation of each vibration or acoustic emission sample, and then standardize it;

[0033] Perform a fast Fourier transform on each standardized sample; that is, append the mean and standard deviation values ​​to the fast Fourier transformed sample to create a feature vector for each sample;

[0034] Based on the range parameters of the training dataset in stage 1, all datasets are additionally scaled to a feature range between 0 and 1; where x is the mean, x~ is the standard deviation, N is the number of sample points, and x i is the i-th sample value; each sample value x i Normalized to z i ; f is the frequency, the formula is as follows:

[0035]

[0036] After preprocessing, the feature vector of each acoustic emission segment finally obtained contains 1014 values, while the feature vector of each vibration segment contains 2045 values;

[0037] Step 3, constructing a fusion classifier including an acoustic emission classifier, a vibration classifier and a combined classifier;

[0038] The feature extraction network of the acoustic emission classifier is a densely connected layer, a first activation layer, and a regularization layer of a stacked structure; the stacked structure is three layers;

[0039] The classification network of the acoustic emission classifier includes: a densely connected layer and a second activation layer;

[0040] The feature extraction network of the vibration classifier is a densely connected layer, a first activation layer, and a regularization layer of a stacked structure; the stacked structure is two layers;

[0041] The classification network of the vibration classifier has the same structure as that of the acoustic emission classifier, which is a cascade of densely connected layers and the second activation layer.

[0042] The feature extraction network of the combined classifier is a cascade of a compressed attention mechanism module and a densely connected layer and a first activation layer of a stacked structure, and the stacked structure is two layers;

[0043] The classification network of the combined classifier is a dense connection layer and a second activation layer, and the dense connection layer of the classification network of the combined classifier is residually connected to the first dense connection layer of the feature extraction network of the combined classifier;

[0044] Residual connections introduce skip paths, which enables the model to be trained efficiently, converge quickly, and improve generalization capabilities; they also make up for the defects of densely connected layers, such as large number of parameters and difficulty in optimization.

[0045] The compressed attention mechanism module (CondensedAttentionNeural Block) consists of two parts: channel attention and spatial attention. Channel attention first extracts information of different features, then renormalizes and calculates weights, multiplies them by different weights, and outputs them through 1x1 convolution fusion; spatial attention captures the correlation of different modal input information in the spatial dimension.

[0046] The first activation layer is the ReakyRelu activation function, and the second activation layer is the SoftMax activation function.

[0047] It also includes: a feature extraction network of a vibration classifier and a classification network of a vibration classifier for predicting leakage status.

[0048] It also includes: a feature extraction network of the acoustic emission classifier and a classification network of the acoustic emission classifier to predict the leakage state.

[0049] Experimental results:

[0050] The confusion matrix is ​​a table layout that supports the visualization of model classification performance and can intuitively express the target value and predicted value. This embodiment evaluates and analyzes the gas pipeline diagnosis results. The confusion matrix is ​​shown in the attached figure. Figure 4 , a single vibration classifier can accurately classify the healthy working condition and slight leakage of gas pipelines, with a diagnostic accuracy of 100%; for moderate leakage, the correct diagnosis rate is 16.33%, and the misdiagnosis rates in slight leakage and severe leakage are 60.33% and 23.14% respectively; for severe leakage, the correct diagnosis rate is 92.50%, and the misdiagnosis rates in healthy working condition, slight leakage and severe leakage are 0.83%, 4.17% and 2.50% respectively.

[0051] For healthy conditions, the correct diagnosis rate of a single acoustic emission classifier is 99.48%, and the misdiagnosis rates for mild leakage, moderate leakage and severe leakage are 0.32%, 0.06% and 0.13% respectively; for mild leakage, the correct diagnosis rate is 89.34%, and the misdiagnosis rates for healthy conditions, moderate leakage and severe leakage are 0.73%, 8.84% and 1.08% respectively; for moderate leakage, the correct diagnosis rate is 41.61%, and the misdiagnosis rates for healthy conditions, mild leakage and severe leakage are 7.90%, 46.61% and 3.89% respectively; for severe leakage, the correct diagnosis rate is 96.09%, and the misdiagnosis rates for healthy conditions, mild leakage and moderate leakage are 1.47%, 1.78% and 0.66% respectively.

[0052] The fusion classifier composed of vibration classifier, acoustic emission classifier and combined classifier can accurately classify the healthy condition and severe leakage of gas pipelines. For mild leakage, the correct diagnosis rate is 90.62%, and the misdiagnosis rate for moderate leakage is 9.38%; for moderate leakage, the correct diagnosis rate is 56.25%, and the misdiagnosis rates for healthy condition, mild leakage and severe leakage are 3.12%, 34.38% and 6.25% respectively.

[0053] like Figure 5The single model and fusion model of the present invention are compared with the existing support vector machine (SVM), XGBoost (Extreme Gradient Boosting), and 1D CNN (1D Convolutional Neural Network); it can be seen that the fusion model of the present invention has the best effect when comprehensive accuracy, detection speed, and model size are considered.

[0054] Based on the above ideal embodiments of the present invention, the relevant staff can make various changes and modifications without departing from the technical concept of the present invention through the above description. The technical scope of the present invention is not limited to the contents of the specification, and its technical scope must be determined according to the scope of the claims.

Claims

1. A pipeline leakage diagnosis method based on a fusion network of acoustic signals and vibration signals, characterized in that: The following steps are involved: Step 1: Collect acoustic emission signals and vibration signals at different knocking angles to construct a pipeline leakage status data set; Step 2: preprocessing the pipeline leakage status data set; Step 3, construct a fusion classifier including an acoustic emission classifier, a vibration classifier and a combined classifier; the feature extraction network of the acoustic emission classifier and the vibration classifier is a dense connection layer, a first activation layer and a regularization layer of a stacked structure; the feature extraction network of the combined classifier is a cascade of a compressed attention mechanism module and a dense connection layer and a first activation layer of a stacked structure, and the classification network of the combined classifier is a cascade of a dense connection layer and a second activation layer; The acoustic signal feature vector, vibration feature vector and knocking angle value are used as inputs of the combined classifier; the feature extraction network of the combined classifier is residually connected to the dense connection layer of its classification network.

2. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: The stacking structure of the feature extraction network of the acoustic emission classifier is 3 layers.

3. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: The stacked structure of the feature extraction network of the vibration classifier is 2 layers.

4. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: The stacked structure of the feature extraction network of the combined classifier is 2 layers.

5. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: Different leakage conditions can be simulated by adjusting the angle between the hammer and the pipeline.

6. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: Leakage status includes: Healthy, Minor Leakage, Moderate Leakage, and Major Leakage.

7. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: Preprocessing includes: Calculate mean, standard deviation, normalization, and fast Fourier transform.

8. The pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network according to claim 1 is characterized in that: The first activation layer is the ReakyRelu activation function, and the second activation layer is the SoftMax activation function.

9. A pipeline leakage diagnosis system based on a fusion network of acoustic signals and vibration signals, characterized in that: include: a memory for storing instructions executable by a processor; A processor, used to execute instructions to implement the pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network as described in any one of claims 1 to 8.

10. A computer readable medium storing computer program code, characterized in that: The computer program code, when executed by a processor, implements the pipeline leakage diagnosis method based on the acoustic signal and vibration signal fusion network as described in any one of claims 1 to 8.

Images