A method for pipeline defect detection using the magnetostrictive characteristics of ultrasonic guided waves

Abstract

The present application provides a method for pipeline defect detection using the magnetostrictive characteristics of ultrasonic guided waves, which relates to the technical field of pipeline defect detection, including: obtaining the geometric and material parameters of multi-layer heterogeneous pipelines, and establishing a mathematical model of the pipeline; according to the mathematical model, calculating the dispersion curve and modal characteristics of ultrasonic guided waves in the pipeline; selecting a suitable excitation frequency and guided wave mode, and optimizing the detection parameters of ultrasonic guided waves; presetting defects of different types and sizes in the pipeline, and analyzing the influence of defects on guided wave propagation characteristics through numerical simulation; building a multi-layer heterogeneous pipeline ultrasonic guided wave detection experimental system, collecting guided wave signals under different defect conditions, and extracting time-frequency domain characteristic parameters. The present invention can realize the effective detection and evaluation of internal defects of multi-layer heterogeneous pipelines at different temperatures, and provides an important technical means for ensuring the safe operation of pipelines.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline defect detection, and in particular to a method for pipeline defect detection using magnetostrictive characteristics of ultrasonic guided waves. Background Art

[0002] How to accurately detect defects inside multi-layer heterogeneous pipeline structures in high-temperature environments is a technical problem that needs to be solved urgently. Current research mainly focuses on the use of magnetostrictive ultrasonic guided wave technology, but it faces multiple challenges. First, the multi-layer heterogeneous pipeline structure is composed of different materials, and the acoustic impedance, sound velocity and other parameters of each layer of materials are significantly different, which leads to complex refraction, reflection and mode conversion of ultrasonic guided waves during propagation, making the dispersion curve complicated and difficult to accurately identify specific modes. Secondly, the high temperature environment has a significant impact on the acoustic properties of the material. For example, the sound velocity changes with temperature, causing the dispersion curve to shift, further increasing the difficulty of mode identification and signal analysis. In addition, the multi-physical field coupling effect, including the interaction of acoustic field, magnetic field, force field and thermal field, makes the waveguide propagation mechanism more complicated. In practical applications, how to effectively suppress the interference of multi-modal guided waves in high-temperature environments, accurately extract specific modal signals carrying defect information, and accurately identify defect types have become key issues. Specifically, in a multi-layer heterogeneous pipeline under high temperature environment, due to the differences in material properties and temperature gradients, multiple modes will be excited during the waveguide propagation process. These modes overlap and interfere with each other, making the echo signal difficult to interpret. At the same time, high temperature will cause changes in the acoustic parameters of the material, causing the dispersion curve to shift, making it difficult to analyze according to the theoretical model at room temperature. In addition, the existence of magnetostrictive effect, thermal expansion effect and structural stress field further aggravates the complexity of the signal. Therefore, how to establish an accurate multi-physics field coupling model according to actual working conditions, predict the propagation characteristics of guided waves in multi-layer heterogeneous pipelines under high temperature environment, and on this basis, develop effective signal processing methods to extract defect information from complex guided wave signals and accurately identify defect types, is the core technical problem currently faced. Summary of the invention

[0003] The present invention provides a method for pipeline defect detection using the magnetostrictive characteristics of ultrasonic guided waves, which mainly includes:

[0004] The geometric and material parameters of the multi-layer heterogeneous pipeline are obtained, and a mathematical model of the pipeline is established; based on the mathematical model, the dispersion curve and modal characteristics of the ultrasonic guided wave in the pipeline are calculated; the appropriate excitation frequency and guided wave mode are selected to optimize the detection parameters of the ultrasonic guided wave; defects of different types and sizes are preset in the pipeline, and the influence of defects on the guided wave propagation characteristics is analyzed through numerical simulation; a multi-layer heterogeneous pipeline ultrasonic guided wave detection experimental system is built, and the guided wave signals under different defect conditions are collected to extract the time-frequency domain characteristic parameters; a guided wave signal feature data set is constructed, and the convolutional neural network, BP neural network and support vector machine models are trained to obtain a defect quantitative identification model; the guided wave signal of the pipeline to be tested is input into the trained defect identification model, and the type, location and size information of the defect are output to realize quantitative non-destructive detection and evaluation of pipeline defects.

[0005] Furthermore, the acquisition of geometric and material parameters of the multi-layer heterogeneous pipeline includes: measuring geometric parameters such as the inner and outer diameters of the pipeline, the thickness of each layer, and other geometric parameters, and obtaining material parameters such as the density, elastic modulus, and Poisson's ratio of each layer of material; and dimensionally processing the geometric and material parameters to eliminate differences in dimensions and numerical ranges of different parameters.

[0006] Furthermore, the calculation of dispersion curves and modal characteristics of ultrasonic guided waves in pipelines includes: solving the wave control equation of the pipeline by combining a semi-analytical finite element method and a transfer matrix method to obtain the phase velocity, group velocity, and displacement and stress modes of the guided waves at different frequencies.

[0007] Furthermore, the optimization of ultrasonic guided wave detection parameters includes: selecting an excitation frequency within a single mode range according to a dispersion curve to avoid multi-mode superposition interference; selecting a low-dispersion, high-attenuation guided wave mode according to modal characteristics to improve detection sensitivity and signal-to-noise ratio.

[0008] Furthermore, the numerical simulation analysis of the influence of defects on the guided wave propagation characteristics includes: introducing defect units of different types, sizes and positions into the pipeline finite element model, using an explicit dynamics algorithm to simulate the interaction between ultrasonic guided waves and defects, extracting the amplitude, frequency and phase information of reflected waves and transmitted waves, and establishing a mapping relationship between defect parameters and guided wave signal characteristics.

[0009] Furthermore, the construction of the multi-layer heterogeneous pipeline ultrasonic guided wave detection experimental system includes: using magnetostrictive sensors as excitation and receiving devices, matching appropriate excitation frequencies and guided wave modes; collecting guided wave signals under different defect conditions, performing preprocessing such as denoising and normalization, and extracting statistical characteristics and energy distribution characteristics in the time and frequency domains.

[0010] Furthermore, the construction of the waveguide signal feature data set includes: extracting characteristic parameters reflecting the defect type and size based on the waveguide signals collected by numerical simulation and experiment, labeling the characteristic data according to the defect parameters, and forming a structured data set for machine learning training.

[0011] Furthermore, the training of convolutional neural network, BP neural network and support vector machine models includes: using data enhancement method to expand training samples, optimizing network structure and model parameters, and improving the accuracy, precision and recall rate of defect recognition; using cross-validation method to evaluate model performance, and selecting the model with the strongest generalization ability as the final defect quantitative recognition model.

[0012] The technical solution provided by the embodiment of the present invention may have the following beneficial effects:

[0013] The present invention discloses a method for pipeline defect detection using the magnetostrictive characteristics of ultrasonic guided waves. The business scenario problem targeted by the method is: how to accurately detect the internal defects of multi-layer heterogeneous pipelines under different temperature environments, especially corrosion pits and cracks, and accurately evaluate their types and sizes. In order to solve this problem, the geometric and material mathematical model of the multi-layer heterogeneous pipeline is first established, and the guided wave dispersion curves at different frequencies are calculated and plotted, so as to optimize the excitation frequency and guided wave mode with smaller dispersion. Then, a multi-physics field coupling finite element model including the acoustic field, magnetic field, force field and temperature field is constructed to simulate the propagation characteristics of guided waves in the pipeline at different temperatures, and analyze the influence of temperature on the guided wave propagation time, amplitude, frequency, displacement and stress distribution. Furthermore, by presetting defects of different types and sizes in the pipeline model, the influence of defects on the guided wave signal is studied. Based on the above theory and simulation results, an ultrasonic guided wave detection experimental system is built, and a magnetostrictive transducer is used to excite and receive the guided wave signal, and the characteristic parameters such as the propagation time, amplitude and frequency of the signal are collected. Finally, a waveguide signal feature dataset containing different defect types and sizes was constructed, and a convolutional neural network model was used for training and testing to achieve high-precision identification of defect types and sizes. This method can effectively detect and evaluate internal defects of multi-layer heterogeneous pipelines at different temperatures, providing an important technical means to ensure the safe operation of pipelines. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 The present invention is a flow chart of a method for pipeline defect detection using the magnetostrictive characteristics of ultrasonic guided waves. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] like Figure 1 In this embodiment, a method for pipeline defect detection using magnetostrictive characteristics of ultrasonic guided waves may specifically include:

[0017] S101. Establish a geometric and material mathematical model of a multi-layer heterogeneous pipeline structure, wherein the geometric parameters include the inner and outer diameters of the pipeline and the thickness of each layer, and the material parameters include the density ρ, Young's modulus E and Poisson's ratio ν of each layer of material.

[0018] Based on the actual measurement results of the pipeline, the structural information of the pipeline is obtained, including the inner and outer diameters of the pipeline and the thickness of each layer of material. The material properties of the pipeline are obtained, including the density, Young's modulus, and Poisson's ratio of each layer of material. The initial data of the obtained pipeline geometry and material parameters are Z-score standardized, and the calculation formula is (x-μ) / σ, where x is the original data, μ is the average value of the data, and σ is the standard deviation of the data. The data of different dimensions are converted into a unified dimensionless form to obtain the normalized geometric parameters and material parameters. If the normalized geometric parameters and material parameters meet the preset range, they are used as input parameters of the finite element model. In the finite element analysis software COMSOL, a geometric model of a multi-layer heterogeneous pipeline structure is created, and the shape of the geometric model is determined by the normalized geometric parameters. In the finite element analysis software COMSOL, according to the normalized material parameters, material properties are assigned to each part of the geometric model, including the density, Young's modulus, and Poisson's ratio of each layer of material, to determine the material properties of the finite element model. The boundary conditions for ultrasonic guided wave propagation are set, including the constraints at both ends of the pipeline and the mode of action of the excitation signal. The excitation signal uses a sine pulse signal modulated by a Hanning window to determine the propagation path of the ultrasonic guided wave in the model. In the finite element analysis software COMSOL, ultrasonic guided wave excitations of different frequencies are applied to the model, and the frequency range is determined according to the actual application requirements. In the finite element analysis software COMSOL, the propagation velocity and wavelength of the guided wave at different frequencies are calculated to obtain the original data of the dispersion curve. According to the original data of the dispersion curve, the cutoff frequency and frequency-wave number relationship of different modal guided waves are analyzed to determine the propagation characteristics of different modal guided waves. The displacement field and stress field distribution of different modal guided waves at different frequencies are analyzed using the finite element analysis software COMSOL. The rate of change of waveguide energy with propagation distance at each frequency is calculated to obtain the energy distribution law of guided wave propagation in multi-layer heterogeneous pipeline structures.

[0019] Specifically, in the actual pipeline measurement process, an ultrasonic thickness gauge can be used to obtain the inner and outer diameters of the pipeline and the thickness of each layer of material. For example, a certain oil pipeline, through field measurement, has an outer diameter of 500 mm, an inner diameter of 480 mm, and a wall thickness of 10 mm, including an anti-corrosion coating with a thickness of 2 mm. At the same time, it is necessary to obtain the properties of each layer of the pipeline material. For example, the main material of the pipeline is steel, with a density of 7850 kg per cubic meter, a Young's modulus of 200 GPa, and a Poisson's ratio of 0.3; the anti-corrosion coating material is polyethylene, with a density of 950 kg per cubic meter, a Young's modulus of 1 GPa, and a Poisson's ratio of 0.4. The obtained pipeline geometry and material parameters are Z-score standardized. The purpose of standardization is to eliminate the differences in dimensions and numerical ranges between different parameters, so that different parameters have the same weight in subsequent analysis. For example, the geometric parameters of the above pipeline are standardized. Assuming that the average value of the outer diameter of multiple similar pipelines is 450 mm and the standard deviation is 50 mm, the standardized value of the outer diameter of the pipeline is (500-450) / 50=1. Similarly, the standardized values ​​of other geometric parameters and material parameters can be calculated. After standardization, all parameters will be converted into dimensionless values, and the mean value is close to 0 and the standard deviation is close to 1. The advantage of this is that it can avoid some parameters with large values ​​from dominating in subsequent analysis, thereby improving the accuracy and reliability of the model. If the standardized parameters meet the preset range, for example, the standardized value of the outer diameter is between -2 and 2, and the standardized value of the wall thickness is between -1 and 1, they can be used as input parameters of the finite element model. In the finite element analysis software, it is first necessary to create a geometric model of a multi-layer heterogeneous pipeline structure. For example, in COMSOL, a cylinder can be created to represent the main part of the pipeline. Its radius is determined by the normalized outer diameter parameter, and the height can be set according to actual needs. Then, create a coaxial cylinder to represent the internal cavity, and its radius is determined by the normalized inner diameter parameter. Finally, create a thin layer cylinder to represent the anti-corrosion coating, and its thickness is determined by the normalized coating thickness parameter. In this way, a pipeline model containing three layers of different materials is constructed. In COMSOL, material properties need to be assigned to each part of the geometric model. For example, the material of the main part can be set to steel, and the density, Young's modulus, and Poisson's ratio can be set to standardized values ​​of 7850 kg per cubic meter, 200 GPa, and 0.3 respectively. Set the material of the internal cavity to air, and its material properties can be ignored. Set the material of the anti-corrosion coating part to polyethylene, and set the density, Young's modulus, and Poisson's ratio to standardized values ​​of 950 kg per cubic meter, 1 GPa, and 0.4 respectively. This completes the material property setting of the finite element model. Next, it is necessary to set the boundary conditions for the propagation of ultrasonic guided waves. For example, the displacement at one end of the pipeline can be fixed, and a Hanning window modulated sinusoidal pulse signal can be applied to the other end as excitation.The frequency of the excitation signal can be determined according to the actual application requirements. For example, for detecting internal defects in pipelines, a signal with a frequency of 1 MHz can be used. The function of the Hanning window is to make the amplitude of the excitation signal rise smoothly from zero to the maximum value, and then smoothly drop to zero, thereby avoiding numerical calculation problems caused by signal mutations. In this way, the propagation path and excitation mode of ultrasonic guided waves in the model are determined. In COMSOL, ultrasonic guided wave excitations of different frequencies can be applied to the model, for example, from 0.5 MHz to 2 MHz, with a step size of 0.1 MHz. Then, the propagation velocity and wavelength of the guided wave at different frequencies can be calculated. For example, at 1 MHz, the phase velocity of a certain mode of guided wave is calculated to be 3000 meters per second, the group velocity is 2800 meters per second, and the wavelength is 3 mm. Based on these data, a dispersion curve, that is, a curve of the relationship between frequency and wave number, can be drawn. The dispersion curve is an important tool for analyzing the propagation characteristics of guided waves. According to the dispersion curve, the cutoff frequency and frequency-wave number relationship of different mode guided waves can be analyzed. For example, it can be observed that the cutoff frequency of a certain mode of guided wave is 0.8 MHz, which means that below this frequency, the guided wave of this mode cannot propagate. It can also be observed that the slopes of the frequency-wavenumber relationship curves of different mode guided waves are different, which reflects the different dispersion characteristics of different mode guided waves. By analyzing these characteristics, the appropriate guided wave mode and frequency can be selected for detection. Using COMSOL, the displacement field and stress field distribution of different mode guided waves at different frequencies can be analyzed. For example, it can be observed that at 1 MHz, the displacement of a certain mode guided wave is mainly concentrated on the outer surface of the pipe, while the stress is mainly concentrated inside the pipe. This distribution characteristic can be used to guide sensor layout and signal analysis. Finally, the rate of change of waveguide energy with propagation distance at each frequency can be calculated to obtain the energy distribution law of guided wave propagation in a multi-layer heterogeneous pipeline structure. For example, it can be found that during the propagation of a certain mode guided wave, energy gradually leaks from the outer surface of the pipe into the internal fluid, which will cause signal attenuation. By analyzing the energy distribution law, the detection scheme can be optimized to improve the sensitivity and reliability of the detection.

[0020] S102, setting the excitation frequency f of the ultrasonic guided wave, calculating the phase velocity Vp and group velocity Vg of different guided wave modes at the frequency f in combination with material parameters, and drawing a dispersion curve.

[0021] Based on the theoretical analysis of the propagation characteristics of magnetostrictive ultrasonic guided waves in multi-layer heterogeneous pipeline structures, the dispersion curve of ultrasonic guided waves in high temperature environment is obtained. The wave number at different frequencies is obtained by using the matrix eigenvalue solution method, and the frequency-wave number dispersion curve is plotted. The mode of ultrasonic guided waves is determined by combining the dispersion curve and the selected excitation frequency. According to the geometric dimensions and material properties of the multi-layer heterogeneous pipeline structure, including the Young's modulus, Poisson's ratio and density of the inner, outer and middle layer materials, the three-dimensional geometric model and material properties of the pipeline are established in COMSOL. Fixed boundaries and excitation boundaries are set in COMSOL. The excitation boundaries impose displacements corresponding to the selected modes. The propagation of ultrasonic guided waves in the pipeline is calculated by transient analysis. A receiving boundary is set at the other end of the pipeline, and the displacement changes at this position with time are recorded to obtain the received signal. The received signal is subjected to fast Fourier transform to obtain the spectrum of the received signal. Add a temperature field in COMSOL, set the initial temperature and ambient temperature of the pipeline, calculate the temperature distribution of the pipeline through heat conduction analysis, couple the temperature distribution to the structural mechanics module, repeat the previous transient analysis and signal reception process, and obtain the receiving signal of ultrasonic guided waves in a high temperature environment. Make a multi-layer heterogeneous pipeline structure specimen, and process corrosion pits of different sizes on the specimen. Build an experimental system, including a signal generator, a power amplifier, a magnetostrictive transducer, an oscilloscope, and a data acquisition card. The signal generator generates an electrical signal of a specific frequency and amplitude, which is amplified by the power amplifier to drive the magnetostrictive transducer, convert the electrical signal into ultrasonic guided waves, and propagate in the pipeline. At the other end of the pipeline, a magnetostrictive transducer is used to receive the ultrasonic guided wave signal and convert it into an electrical signal. The oscilloscope and data acquisition card collect the electrical signal and transmit it to the computer. Use MATLAB in the computer to perform fast Fourier transform on the collected signal to obtain the signal spectrum. Compare the signal spectrum obtained by COMSOL numerical simulation with the signal spectrum obtained by experimental measurement to verify the accuracy of the numerical model.

[0022] Specifically, the ultrasonic guided wave technology based on the magnetostrictive effect is widely used in non-destructive testing of pipeline structures due to its advantages such as long detection distance and high efficiency. When the pipeline is in a high temperature environment, its material properties and geometric dimensions will change, which will affect the propagation characteristics of the ultrasonic guided wave, especially the dispersion characteristics. Therefore, obtaining the dispersion curve in a high temperature environment is crucial for accurately assessing the safety status of the pipeline structure. The frequency-wavenumber dispersion curve describes the change law of the wave number corresponding to the guided waves of different frequencies when the ultrasonic guided wave propagates in a specific medium. These key data can be obtained by solving the matrix eigenvalues. For example, for a multi-layer pipeline consisting of a steel inner layer, a polyethylene middle layer and a steel outer layer, it is assumed that the wave number of a certain guided wave mode measured at room temperature (25 degrees Celsius) at a frequency of 1 MHz is 2000 per meter, while in a high temperature environment (such as 100 degrees Celsius), due to the change in material properties, the wave number at the same frequency may become 1950 per meter. This change reflects the influence of temperature on the propagation speed of the guided wave, which in turn affects the accuracy of the detection results. After selecting a specific excitation frequency, such as 1.2 MHz, the specific guided wave mode that can be excited at this frequency, such as the longitudinal mode L(0, 2), can be determined based on the known frequency-wave number dispersion curve. The purpose of determining the mode is to be able to specifically excite and receive the guided wave signal of this mode in subsequent numerical simulations and experiments. Using the finite element analysis software COMSOL, a three-dimensional model of a multilayer heterogeneous pipeline can be established. For example, a steel pipeline with an outer diameter of 500 mm, an inner diameter of 480 mm, and a wall thickness of 10 mm is coated with a 2 mm thick polyethylene anti-corrosion layer. In COMSOL, three coaxial cylinders can be created first, representing the steel pipeline, the internal cavity, and the polyethylene coating, and the corresponding geometric dimensions can be assigned. For material properties, for example, the Young's modulus of steel is 200 GPa, the Poisson's ratio is 0.3, and the density is 7850 kg / m3; the Young's modulus of polyethylene is 1 GPa, the Poisson's ratio is 0.4, and the density is 950 kg / m3. In order to simulate the propagation of ultrasonic guided waves, boundary conditions need to be set in the model. For example, one end of the pipeline can be set as a fixed boundary, that is, the displacement of all nodes at this end is zero; the other end is set as an excitation boundary, and the displacement corresponding to the selected mode L(0, 2) is applied. For example, at a frequency of 1.2 MHz, the displacement of the L(0, 2) mode is mainly along the axial direction of the pipeline, so a sinusoidal pulse displacement modulated by a Hanning window along the axial direction can be applied to the excitation boundary. Through transient analysis, the propagation process of ultrasonic guided waves in the pipeline can be calculated, and a receiving boundary is set at the other end of the pipeline to record the displacement change at this position over time. For example, a signal of the axial displacement change over time received at a distance of 1 meter from the excitation boundary can be recorded. The received signal is fast Fourier transformed to obtain its spectrum. For example, it can be observed that there is an obvious peak near 1.2 MHz, which indicates that the received signal mainly contains the guided wave component of this frequency.In order to simulate high temperature environment, a temperature field module can be added to COMSOL. For example, the initial temperature of the pipeline can be set to 25 degrees Celsius and the ambient temperature to 100 degrees Celsius, and the temperature distribution of the pipeline after a period of time can be calculated through heat conduction analysis. For example, it can be calculated that after 30 minutes of heating, the temperature of the outer surface of the pipeline reaches 80 degrees Celsius, while the temperature of the inner surface is 60 degrees Celsius. By coupling the temperature distribution to the structural mechanics module and repeating the previous transient analysis and signal reception process, the received signal of the ultrasonic guided wave in the high temperature environment can be obtained. For example, it can be found that in the high temperature environment, the amplitude of the received signal decreases and the arrival time is delayed, which indicates that the increase in temperature causes the attenuation and speed reduction of the guided wave. In practical applications, the accuracy of the numerical model can be verified by making multi-layer heterogeneous pipeline specimens and processing corrosion defects. For example, a multi-layer pipeline specimen with the same size and material as the numerical model can be made, and corrosion pits of different depths and widths can be processed on its outer surface. A magnetostrictive transducer is used to excite and receive ultrasonic guided wave signals on the specimen. For example, a magnetostrictive transducer with a center frequency of 1.2 MHz can be used to excite the guided wave of the L(0,2) mode, and another identical transducer can be used to receive the signal at a distance of 1 meter from the excitation position. The accuracy of the model can be verified by comparing the signal spectra obtained by numerical simulation and experimental measurement. For example, it can be found that in the presence of corrosion pits, the signal spectra obtained by numerical simulation and experimental measurement both show a decrease in amplitude within a specific frequency range, and the degree of decrease is related to the size of the pits. This consistency shows that the numerical model can accurately predict the propagation characteristics of ultrasonic guided waves in multilayer heterogeneous pipes under high temperature environments, providing important theoretical guidance for practical detection.

[0023] S103. Select an excitation frequency f and a waveguide mode with fewer modes and smaller dispersion according to the dispersion curve.

[0024] Obtain the theoretical model of ultrasonic guided wave propagation in multi-layer heterogeneous pipeline structure, and clarify the meaning of parameters and variables in the model. Based on the theoretical model, MATLAB or Python is used for programming calculation to obtain the dispersion curve of propagation in multi-layer heterogeneous pipeline structure. Analyze the dispersion curve, identify multiple modes and dispersion phenomena, and determine the excitation frequency f with fewer modes and smaller dispersion according to the actual engineering needs and detection accuracy requirements. Combined with the material properties and geometric parameters of the multi-layer heterogeneous pipeline structure, finite element analysis is performed using ANSYS or COMSOL Multiphysics to obtain the guided wave mode at a specific frequency. Through numerical simulation, the propagation of ultrasonic guided waves in multi-layer heterogeneous high-temperature pipelines is obtained, and the time domain signal of the reflected echo is recorded. Analyze the characteristics of the reflected echo, quantify the time delay, amplitude change and frequency change, and if there is a significant change, it is judged that there is a defect in the pipeline. The reflected echo signal is denoised by wavelet transform to obtain the denoised signal data. Based on the denoised signal data, short-time Fourier transform (STFT) is used for time-frequency analysis to obtain the time-frequency distribution diagram of the signal, and the time-frequency characteristics of different types of defects are analyzed. Obtain pipeline defect data. If the data already has time-frequency feature labels, use data enhancement methods such as rotation, flipping, and noise addition to build an expanded defect data set. Based on the expanded defect data set, use convolutional neural network (CNN) and support vector machine (SVM) for model training, and set appropriate network structure and parameters. Use an independent test set to test the training model, and determine the optimal defect type recognition model through comparative analysis of indicators such as accuracy and recall rate.

[0025] Specifically, the propagation of ultrasonic guided waves in multi-layer heterogeneous pipe structures can be described by theoretical models. Taking a three-layer cylindrical tube as an example, the inner and outer layers are steel and the middle layer is polyethylene. The key parameters in the model include the thickness, density, Young's modulus and Poisson's ratio of each layer. The variables include frequency and wave number. The dispersion curve can be obtained by solving the characteristic equation of the model using MATLAB programming. The dispersion curve shows the change of phase velocity or group velocity with frequency. The analysis curve can identify multiple modes, such as longitudinal L(0,1), L(0,2) and torsional T(0,1). The low frequency band usually has fewer modes and smaller dispersion, which is suitable for actual detection. For example, for a three-layer tube with an outer diameter of 200mm and a wall thickness of 10mm, there may be only two modes, L(0,1) and T(0,1), in the range of 100-200kHz. Among them, L(0,1) has the smallest dispersion at around 150kHz and can be selected as the excitation frequency. After determining the excitation frequency, COMSOL can be used for finite element analysis. Establish a three-dimensional model, set material parameters and boundary conditions. For example, apply 150kHz axial displacement excitation at one end of the pipeline to simulate the L(0,1) mode. The analysis results can show the stress and strain distribution and verify the mode shape. The numerical simulation of guided wave propagation in high-temperature pipelines needs to consider the effect of temperature on material parameters. For example, the Young's modulus of steel decreases with increasing temperature and may decrease by 5% at 200°C. The simulation results may show that under the same excitation, the reflected echo in the high-temperature environment has a time delay of 0.1ms compared with the normal temperature, the amplitude decreases by 10%, and the main frequency decreases from 150kHz to 145kHz. These changes may indicate defects in the pipeline, such as corrosion thinning. In order to improve the signal-to-noise ratio, the echo signal can be denoised by wavelet. Selecting the appropriate wavelet basis function (such as db4) and the number of decomposition layers (such as 3 layers) can effectively remove high-frequency noise and retain useful information. The denoised signal can be used for subsequent analysis. Short-time Fourier transform (STFT) can reveal the time-frequency characteristics of the signal. Perform STFT on the denoised signal, select the appropriate window function (such as Hanning window) and window length (such as 1ms), and obtain the time-frequency distribution diagram. Different types of defects show different characteristics in the diagram. For example, local corrosion may show broadband energy at a specific time point, while cracks may show narrowband energy with a longer duration. When constructing a defect dataset, data enhancement technology can be used. For the existing time-frequency feature map, operations such as ±5° rotation, horizontal flipping, and adding 5% Gaussian noise can be performed. This can expand the original dataset by 5-10 times and improve the generalization ability of the model. Based on the expanded dataset, convolutional neural network (CNN) and support vector machine (SVM) models can be trained. The CNN structure can contain 3-4 convolutional layers, each followed by a maximum pooling layer and finally a fully connected layer. SVM can use a radial basis function (RBF) kernel. The optimal parameters, such as the convolution kernel size and learning rate of CNN, the penalty factor C and kernel parameter γ of SVM, are determined by cross-validation. The model performance is evaluated using an independent test set.Assume that there are three categories: corrosion, cracks, and no defects, with 100 samples each. The CNN model may achieve an overall accuracy of 95%, with a corrosion recognition rate of 98%, a crack recognition rate of 92%, and a no defect recognition rate of 95%. The SVM model may have an overall accuracy of 92%, with recognition rates of 95%, 90%, and 91% for each category, respectively. Considering the accuracy, recall rate, and F1 score, the CNN model may be better. This method based on ultrasonic guided waves and machine learning provides an effective means for non-destructive testing of high-temperature multi-layer heterogeneous pipelines. Through theoretical modeling, numerical simulation, and experimental verification, the propagation characteristics of guided waves can be accurately obtained. Combined with advanced signal processing and artificial intelligence technology, it can realize the automatic identification and positioning of pipeline defects, and improve detection efficiency and reliability. This is of great significance for ensuring the safe operation of industrial equipment and extending its service life.

[0026] S104. A finite element model is established by combining the structural parameters and material parameters of the multi-layer heterogeneous pipeline. The model includes acoustic field, magnetic field, force field and temperature field modules.

[0027] According to the material parameters and structural parameters of multi-layer heterogeneous pipelines, a multi-physics field coupling finite element model including acoustic field, magnetic field, force field and temperature field is established, and appropriate boundary conditions and initial conditions are applied to simulate the actual working conditions of multi-layer heterogeneous pipelines. Through finite element simulation, the propagation characteristic data of ultrasonic guided waves in multi-layer heterogeneous pipelines at different frequencies are obtained. Fast Fourier transform is used to process the simulation data, and dispersion curves are drawn to analyze the dispersion characteristics of ultrasonic guided waves in multi-layer heterogeneous pipelines. The dispersion curves are modally analyzed using short-time Fourier transform to obtain the propagation characteristics of ultrasonic guided waves under different modes. According to the analysis results of dispersion characteristics and modal characteristics, appropriate excitation frequency and guided wave mode are selected to achieve the best detection effect. The displacement field distribution of ultrasonic guided waves when propagating in multi-layer heterogeneous pipelines is obtained through simulation, and the energy attenuation law of guided waves in pipelines is analyzed. The analysis results of dispersion characteristics, multi-modal characteristics, energy attenuation characteristics, etc. are combined to optimize the ultrasonic guided wave detection parameters, laying the foundation for subsequent pipeline defect detection. In actual detection, the ultrasonic transducer array is arranged reasonably to stimulate the optimized guided wave mode and receive the ultrasonic guided wave signal propagating in the pipeline. The characteristic parameters of the guided wave signal are extracted by using signal processing methods such as wavelet transform, and the presence of defects in the pipeline is determined by the change of characteristic parameters. Further quantitative analysis is performed on the suspected defect area to determine the type, location, size and other information of the defect, providing a basis for pipeline safety assessment and maintenance decision-making.

[0028] Specifically, ultrasonic guided wave detection of multi-layer heterogeneous pipelines involves complex multi-physical field coupling problems. Taking a three-layer composite pipeline as an example, the inner and outer layers are metals, and the middle is polymer. When establishing a finite element model, the interaction between the acoustic field, magnetic field, force field and temperature field needs to be considered. The acoustic field describes the propagation of ultrasonic waves, the magnetic field reflects the magnetism of the material, the force field reflects the stress and strain of the pipeline, and the temperature field affects the material parameters. For example, at a high temperature of 200°C, the Young's modulus of steel may decrease by 5%, resulting in a decrease in wave velocity. In terms of boundary conditions, axial displacement excitation can be applied to the pipe end to simulate transducer excitation; the outer surface is set as a free boundary, and the inner surface needs to consider coupling with the fluid. After the time domain data obtained by simulation is fast Fourier transformed, a dispersion curve can be drawn. Taking a three-layer pipe with an outer diameter of 200mm and a wall thickness of 10mm as an example, multiple modes may appear in the range of 0-500kHz, such as L(0,1), L(0,2), T(0,1), etc. The dispersion curve shows the change of phase velocity or group velocity with frequency. The low frequency band usually has a small dispersion and is suitable for actual detection. Short-time Fourier transform can further analyze the energy distribution of each mode. For example, in the range of 100-200kHz, the L(0,1) mode may be dominant, and its group velocity reaches a maximum value of 3000m / s at around 150kHz. According to the results of dispersion analysis, the appropriate excitation frequency and mode can be selected. For example, if the 150kHz L(0,1) mode is selected, its wavelength is about 20mm, which is smaller than the pipe wall thickness, which is conducive to defect detection. The simulation results can show the propagation process of guided waves in the pipeline. Taking the axial displacement as an example, it may show periodic changes, and the displacement at the peak reaches the micron level. In terms of energy attenuation, due to material damping and geometric diffusion, the amplitude may attenuate by 20% for every 1m of propagation. When optimizing the detection parameters, it is necessary to weigh the sensitivity and propagation distance. Higher frequencies are conducive to small defect detection, but the attenuation is faster. For example, the 300kHz L(0,2) mode has high sensitivity, but the propagation distance may be only half of the 100kHz L(0,1) mode. In actual detection, a multi-frequency excitation strategy can be adopted, such as using 100kHz and 300kHz at the same time, taking into account the detection range and accuracy. The arrangement of the transducer array needs to consider beamforming. Taking the 8-element ring array as an example, directional excitation can be achieved by adjusting the phase difference of each element. If you want to excite the L(0,1) mode propagating in the axial forward direction, the phase difference of the excitation signal of the adjacent elements can be 45°. When receiving, the delay-sum algorithm is used to process the array signal to improve the signal-to-noise ratio. In terms of signal processing, wavelet transform is suitable for extracting non-stationary signal features. Selecting appropriate wavelet basis functions (such as db4) and decomposition layers (such as 3 layers) can effectively separate useful signals and noise. Characteristic parameters may include amplitude, arrival time, main frequency, etc. For example, the reflected signal at the defect may arrive 0.1ms earlier than the healthy part, the amplitude is reduced by 30%, and the main frequency is reduced from 150kHz to 140kHz. Quantitative analysis of defects can be combined with theoretical models and databases. Taking local corrosion as an example, the reflection coefficient is related to the defect depth and circumferential range.By comparing the measured data with the pre-calculated theoretical value, the defect size can be estimated. For example, a reflection coefficient of 0.2 may correspond to a corrosion pit with a depth of 30% of the wall thickness and a circumferential range of 60°. Combining the test results of multiple frequencies and modes can improve the evaluation accuracy. This information provides an important basis for pipeline safety assessment, helps to formulate a reasonable maintenance plan, extend the service life of equipment, and reduce operational risks.

[0029] S105, setting the selected excitation frequency f and waveguide mode, applying corresponding boundary conditions, and performing multi-physical field coupling numerical calculations.

[0030] According to the material properties and geometric dimensions of the multi-layer heterogeneous pipeline, a three-dimensional solid model of the pipeline is established, including the different material property parameters of the inner layer, outer layer and middle layer of the pipeline. The finite element analysis software ANSYS is used to mesh the pipeline model and generate a high-quality finite element model. Based on the actual working conditions of the pipeline, the boundary conditions and load conditions of the pipeline are set. Through modal analysis, the natural frequency and vibration mode of the pipeline are obtained, and the appropriate excitation frequency range is selected to avoid the influence of dispersion effect on the detection results. Harmonic response analysis is performed, and the selected excitation frequency is applied to obtain the propagation characteristics of ultrasonic guided waves in the pipeline at different frequencies, including displacement cloud map, stress cloud map and energy distribution cloud map. According to the results of harmonic response analysis, the propagation path, energy accumulation area and attenuation law of guided waves in the pipeline are obtained, providing a theoretical basis for defect detection. Through time history analysis, the propagation process of ultrasonic guided waves in the pipeline is simulated, the change law of guided wave signals in the time domain and frequency domain is obtained, the propagation ability and sensitivity of guided waves of different modes in the pipeline are judged, and the optimal guided wave mode is selected. Based on the above analysis results, the dispersion and multimodal characteristics of ultrasonic guided waves propagating in multi-layer heterogeneous high-temperature pipelines are obtained, and the appropriate excitation method, excitation position and sensor layout scheme are determined, laying the foundation for pipeline defect detection.

[0031] Specifically, the ultrasonic guided wave detection modeling of multi-layer heterogeneous pipelines is a complex process that requires consideration of multiple factors. First, it is crucial to establish a three-dimensional solid model based on the actual structure of the pipeline. Taking a common three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. When meshing in ANSYS, it is necessary to balance the calculation accuracy and efficiency. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, which meets the calculation accuracy requirements. In terms of boundary condition setting, axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. Modal analysis is a key step in determining the appropriate excitation frequency. Through calculation, it may be found that the pipeline has multiple modes in the range of 0-500kHz, such as L(0,1), L(0,2), T(0,1), etc. Among them, the L(0,1) mode has a small dispersion in the low frequency band (such as 50-150kHz), which is suitable for actual detection. Selecting 100kHz as the excitation frequency can avoid serious dispersion effects while ensuring sufficient detection sensitivity. Harmonic response analysis can reveal the propagation characteristics of guided waves in the pipeline. Taking the 100kHz L(0,1) mode as an example, the displacement cloud map may show that the axial displacement changes periodically, and the maximum displacement reaches the micron level. The energy distribution cloud map may show that the energy is mainly concentrated in the middle of the pipe wall, which is conducive to detecting defects in the middle of the pipeline. The stress cloud map may show that the maximum stress appears on the inner and outer surfaces of the pipe, which suggests that fatigue cracks may be more likely to occur in these areas. Time history analysis can simulate the propagation process of guided waves. For example, a 100kHz sine wave packet with 5 cycles is excited and its propagation in the pipeline is observed. It may be found that the wave packet gradually broadens during propagation, which is caused by slightly different propagation speeds of different frequency components. Through fast Fourier transform, the spectrum of the signal can be obtained, thereby judging the energy distribution of each mode. For example, it may be found that near 100kHz, the L(0,1) mode is dominant, while the energy of other modes such as L(0,2) is weaker. Based on the above analysis, the optimal detection scheme can be determined. For example, the 100kHz L(0,1) mode is selected as the main detection mode, and a ring transducer array is used for excitation and reception. The excitation position can be selected at a distance of about 2m from the area to be detected, so as to avoid the influence of the near-field effect. The sensor can be arranged on the opposite side of the excitation position to form a transmission detection configuration. At the same time, a sensor can also be arranged on the excitation side to receive the reflected signal.The advantages of this detection scheme are: first, the L(0,1) mode has small dispersion in the low frequency band, which is conducive to signal interpretation; second, the wavelength of 100kHz is about 30mm, which is smaller than the pipe wall thickness, which is conducive to the detection of smaller defects; third, the annular array can achieve full circumferential excitation and reception, improving the coverage of detection. However, this scheme also has limitations. For example, the sensitivity to axial cracks may not be as good as the T(0,1) mode. Therefore, in practical applications, it may be necessary to combine multiple modes and frequencies for comprehensive detection to improve the reliability and comprehensiveness of detection.

[0032] S106. Obtain the propagation time t, amplitude A and frequency f' of the first waveguide signal at different positions in the calculation results, and draw time domain and frequency domain spectra of the first waveguide signal.

[0033] According to the material properties and geometric dimensions of the multi-layer heterogeneous pipeline, a three-dimensional finite element model of the pipeline is established, and the boundary conditions and initial conditions are set reasonably. The finite element method is used to mesh the pipeline model, and the continuous pipeline structure is discretized into a finite number of units and nodes. By solving the wave equation in the pipeline, the propagation characteristics of ultrasonic guided waves in the pipeline at different frequencies are calculated, including parameters such as phase velocity, group velocity and wavelength. The dispersion curve of ultrasonic guided waves is obtained, and the relationship between frequency and phase velocity and group velocity is analyzed to select a suitable excitation frequency range for subsequent defect detection. Based on the analysis results of the dispersion curve, the displacement field and stress field distribution of ultrasonic guided waves in the pipeline under different modes are calculated. By comparing the displacement and stress amplitudes of each mode, the sensitivity of different modes to defects is judged, and the dominant mode and its sensitive area are determined. Defects of different types, sizes and locations, such as cracks, corrosion and thinning, are introduced into the three-dimensional finite element model of the pipeline. The interaction between ultrasonic guided waves and defects is simulated by numerical simulation, and the guided wave signals at different positions of the pipeline are obtained. The simulated guided wave signals are subjected to wavelet transform to extract the energy distribution characteristics of the signal in the time-frequency domain. The amplitude, frequency and phase information of the wavelet coefficients are used to construct defect-sensitive feature vectors. The support vector machine (SVM) algorithm is used to classify and regress the extracted feature vectors. By training known defect samples, the mapping relationship between defect type, location and size and waveguide signal characteristics is established. The trained SVM model is used to identify and quantitatively evaluate unknown defects. The results of defect identification and quantitative evaluation are comprehensively analyzed with the design parameters and service conditions of the pipeline to evaluate the health status and remaining life of the pipeline. According to the evaluation results, targeted inspection and maintenance strategies are formulated to ensure the safe operation of the pipeline.

[0034] Specifically, the ultrasonic guided wave detection modeling of multi-layer heterogeneous pipelines is a complex process that requires consideration of multiple factors. First, it is crucial to establish a three-dimensional solid model based on the actual structure of the pipeline. Taking a common three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. When meshing in finite element analysis software, it is necessary to balance the calculation accuracy and efficiency. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, which meets the calculation accuracy requirements. In terms of boundary condition setting, axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. By solving the wave equation, the propagation characteristics of ultrasonic guided waves in the pipeline can be obtained. For example, for the L(0,1) mode, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. The analysis of the dispersion curve is crucial for selecting the appropriate excitation frequency. By comparing the dispersion curves of different modes, it can be found that the L(0,1) mode has a small dispersion in the low frequency band (such as 50-150kHz), which is suitable for actual detection. Choosing 100kHz as the excitation frequency can avoid serious dispersion effects while ensuring sufficient detection sensitivity. By calculating the displacement field and stress field distribution under different modes, the sensitivity of each mode to defects can be determined. For example, for the L(0,1) mode, the axial displacement is the largest in the middle of the pipe wall, which means that this mode is most sensitive to defects in the middle of the pipe wall. The tangential displacement of the T(0,1) mode is the largest on the surface of the pipe wall, so it is more sensitive to surface defects. After introducing defects into the model, the interaction between ultrasonic guided waves and defects can be simulated. For example, an axial crack with a depth of 30% of the wall thickness and a length of 10 mm is introduced at a distance of 2 m from the excitation source. Through simulation, it can be found that when the L(0,1) modal guided wave encounters the crack, an obvious reflection wave will be generated, and the reflection coefficient is about 10%. At the same time, the amplitude of the transmitted wave will decrease and the waveform will be distorted. The wavelet transform of the simulated guided wave signal can extract the energy distribution characteristics in the time and frequency domain. For example, using Morlet wavelet to decompose the signal, it may be found that the energy of the defect reflection signal in the 80-120kHz frequency band is significantly enhanced, and a peak corresponding to the defect position appears on the time axis. These features can be used to construct defect-sensitive feature vectors. The support vector machine (SVM) algorithm is used for defect identification and quantitative evaluation.By training known defect samples, a mapping relationship between defect features and guided wave signal features can be established. For example, for axial cracks of different depths (10%, 20%, 30% wall thickness), it may be found that the reflection coefficient is approximately linearly related to the defect depth. Using this relationship, the estimation of unknown defect depth can be achieved. Finally, the results of defect identification and quantitative evaluation are comprehensively analyzed with the design parameters and service conditions of the pipeline to evaluate the health status and remaining life of the pipeline. For example, if the maximum defect depth detected is 40% of the wall thickness, according to the pipeline design specifications, it may need to be repaired within the next 6 months. Based on these evaluation results, targeted inspection and maintenance strategies can be formulated, such as increasing the inspection frequency, local repair or replacement of pipe sections, etc., to ensure the safe operation of the pipeline.

[0035] S107. Analyze the variation of the propagation time t, amplitude A and frequency f' of the first waveguide signal at different temperatures T, as well as the displacement u and stress σ distribution of the waveguide mode at different temperatures T, to obtain the influence of temperature T on the waveguide propagation characteristics.

[0036] According to the material properties and geometric dimensions of the multi-layer heterogeneous pipeline, a mathematical model of the pipeline is established using finite element software such as COMSOL, including different material property parameters of the inner layer, outer layer and middle layer of the pipeline. The pipeline model is meshed to obtain a finite element mesh model, and appropriate boundary conditions and initial conditions are set. The appropriate ultrasonic guided wave excitation signal and frequency are selected, and the propagation process of ultrasonic guided waves in the pipeline is simulated by finite element simulation to obtain the guided wave signals at different positions. MATLAB is used to perform time-frequency analysis on the guided wave signal to obtain the dispersion curves of different modal guided waves and determine the relationship between the group velocity and phase velocity of each mode and the frequency. The temperature parameters of the pipeline are changed, and the guided wave propagation simulation and dispersion analysis are repeated to obtain the change data of the propagation time, amplitude and frequency of the guided wave signal at different temperatures. The least squares method is used to curve fit the guided wave propagation parameters at different temperatures to obtain the functional relationship between temperature and these parameters. At different temperatures, the displacement and stress data of each mode in the guided wave signal are extracted, and the distribution cloud diagram of displacement and stress is drawn using visualization software such as ParaView to qualitatively analyze the influence of temperature on the displacement and stress distribution of the guided wave mode. According to the amplitude change data of the guided wave signal, the exponential fitting method is used to calculate the attenuation coefficient of the guided wave at different temperatures, and the influence of temperature on the attenuation of the guided wave energy is quantitatively analyzed. The influence of temperature on the guided wave propagation time, amplitude, frequency, modal displacement stress distribution and attenuation is comprehensively analyzed, and a quantitative relationship model between temperature and guided wave propagation characteristic parameters is established. For example, by using methods such as multivariate regression analysis, the quantitative relationship between temperature and ultrasonic guided wave propagation characteristics in multi-layer heterogeneous pipelines is obtained, laying the foundation for subsequent defect detection.

[0037] Specifically, the ultrasonic guided wave detection modeling of multi-layer heterogeneous pipelines is a complex process that requires consideration of multiple factors. First, it is crucial to establish a three-dimensional solid model based on the actual structure of the pipeline. Taking a common three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. When meshing in finite element analysis software, it is necessary to balance the calculation accuracy and efficiency. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, which meets the calculation accuracy requirements. In terms of boundary condition setting, axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. By solving the wave equation, the propagation characteristics of ultrasonic guided waves in the pipeline can be obtained. For example, for the L(0,1) mode, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. The analysis of the dispersion curve is crucial for selecting the appropriate excitation frequency. By comparing the dispersion curves of different modes, it can be found that the L(0,1) mode has a small dispersion in the low frequency band (such as 50-150kHz), which is suitable for actual detection. Choosing 100kHz as the excitation frequency can avoid serious dispersion effects while ensuring sufficient detection sensitivity. By calculating the displacement field and stress field distribution under different modes, the sensitivity of each mode to defects can be determined. For example, for the L(0,1) mode, the axial displacement is the largest in the middle of the pipe wall, which means that this mode is most sensitive to defects in the middle of the pipe wall. The tangential displacement of the T(0,1) mode is the largest on the surface of the pipe wall, so it is more sensitive to surface defects. The effect of temperature on guided wave propagation characteristics is an important research direction. By changing the temperature parameters in the model, the propagation behavior of guided waves at different temperatures can be simulated. For example, when the temperature increases from 20°C to 100°C, the elastic modulus of the pipe material may decrease by 5% to 1%.

[0038] 0%, which will lead to a decrease in the guided wave velocity. Specifically, the group velocity of the L(0,1) mode may decrease from 3000m / s to 2900m / s. At the same time, the increase in temperature will increase the acoustic impedance of the material, resulting in an increase in the attenuation of the guided wave signal amplitude. The least squares method is used to curve fit the guided wave propagation parameters at different temperatures, and the functional relationship between temperature and these parameters can be obtained. For example, there may be the following relationship between the guided wave velocity v and the temperature T: v = v0 (1-αT), where v0 is the velocity at the reference temperature and α is the temperature coefficient. By fitting the experimental data, it is possible to obtain the order of α≈10-4 / ℃. Using visualization software such as ParaView to draw the distribution cloud map of displacement and stress, the effect of temperature on the guided wave mode distribution can be intuitively observed. For example, at high temperatures, the energy of the L(0,1) mode may be more concentrated on the outer surface of the pipe, which means that the sensitivity of this mode to outer surface defects at high temperatures may increase. The exponential fitting method is used to calculate the attenuation coefficient of the guided wave at different temperatures, and the influence of temperature on the attenuation of the guided wave energy can be quantitatively analyzed. For example, the attenuation coefficient α may have the following relationship with the temperature T: α = α0exp(βT), where α0 is the attenuation coefficient at the reference temperature and β is the temperature sensitivity coefficient. By fitting the experimental data, it is possible to obtain β≈0.01 / ℃. By comprehensively analyzing the influence of temperature on the waveguide propagation time, amplitude, frequency, modal displacement stress distribution and attenuation, a quantitative relationship model between temperature and waveguide propagation characteristic parameters can be established. For example, using the multivariate regression analysis method, a model of the following form can be obtained: Y = a0 + a1T + a2T2 + a3f + a4fT, where Y can be a parameter such as waveguide velocity, amplitude or attenuation coefficient, T is temperature, f is frequency, and a0~a4 are regression coefficients. This model can be used to predict the waveguide propagation characteristics under different temperature and frequency conditions, and provide a theoretical basis for temperature compensation in actual detection.

[0039] Defects of different types and sizes are preset in the pipeline. The defect types include corrosion pits and cracks. The size parameters include defect depth h and width w. Combined with the influence of temperature T on the guided wave propagation characteristics, the multi-physical field coupling numerical calculation is repeated to obtain the propagation time t, amplitude A and frequency f of the second guided wave signal in the defective pipeline.

[0040] According to the material properties and geometric dimensions of the multi-layer heterogeneous pipeline, a three-dimensional solid model of the pipeline is established, including the material parameters of the inner layer, outer layer and middle layer of the pipeline. The finite element method is used to discretize the pipeline solid model in space and generate a finite element mesh model containing nodes and units. Defects of different types and sizes are preset in the finite element mesh model, including corrosion pits and cracks. The defect parameters include depth h and width w, and the introduction of defects is achieved by adjusting the coordinates of the mesh nodes at the defect position. According to the actual working conditions of the pipeline, the mechanical boundary conditions and initial conditions of the pipeline model are set. An ultrasonic guided wave excitation signal is applied to one end of the pipeline model, and the excitation signal can be a narrow-band modulated sine pulse. Considering the influence of temperature T on the material properties of the pipeline, the thermal-acoustic-solid multi-physics field coupling finite element method is used to solve the propagation process of ultrasonic guided waves in defective multi-layer heterogeneous pipelines. The time domain waveform of the guided wave signal at different positions in the pipeline model is extracted, and the characteristic parameters of the guided wave signal such as propagation time t, amplitude A and main frequency f are obtained by using signal processing methods such as fast Fourier transform. Based on the dispersion equation, the dispersion curves of different modal guided waves in the pipeline are calculated, that is, the relationship curves between phase velocity, group velocity and frequency. Through modal decomposition, the displacement and stress distribution of the guided wave signal on the pipeline cross section are obtained, and the area where the guided wave energy is mainly concentrated is determined. Comprehensively considering the dispersion characteristics and modal distribution, the guided wave modes and frequencies with high sensitivity and low attenuation are selected as detection parameters. The characteristic parameters such as propagation time, amplitude and frequency of the guided wave signals in defective pipelines and intact pipelines are compared. The quantitative relationship between the defect parameters (type, location, depth, width) and the characteristic parameters of the guided wave signals is established using mathematical statistics methods, and a machine learning model for quantitative defect evaluation is constructed. The model parameters are optimized through training. Using the trained machine learning model, the type, location and size of the defect are inverted according to the characteristic parameters of the guided wave signal, and the confidence of the evaluation is given. By comparing with the results of non-destructive testing methods such as ray ultrasound and eddy current commonly used in the field of pipeline detection, the effectiveness and reliability of the established defect quantitative evaluation method are verified, and the error level of the evaluation results is analyzed. Under the actual operation conditions of the pipeline, the guided wave signals at different temperatures are obtained, the influence of temperature on the propagation velocity and attenuation of the guided wave is studied, and a quantitative compensation model of temperature T and guided wave characteristic parameters is established to achieve temperature compensation calibration of the defect assessment results. Finally, by arranging a sensor array in the actual pipeline and collecting guided wave signals under different defect conditions, the constructed multi-layer heterogeneous pipeline defect quantitative assessment method is verified in engineering applications, and the assessment model and detection scheme are optimized and improved to form a complete set of pipeline non-destructive testing and health assessment technology systems to guide the safe operation and maintenance of pipelines.

[0041] Specifically, the modeling of ultrasonic guided wave detection of multi-layer heterogeneous pipelines is a complex process that requires consideration of multiple factors. First, it is crucial to establish a three-dimensional solid model based on the actual structure of the pipeline. Taking a common three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. When meshing in finite element analysis software, it is necessary to balance the calculation accuracy and efficiency. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, which meets the calculation accuracy requirements. When presetting defects in the mesh model, corrosion pits and cracks of different types and sizes can be considered. For example, a corrosion pit with a depth of 30% of the wall thickness and a width of 20mm, or a crack with a length of 10mm and a width of 0.5mm. In terms of boundary condition setting, an axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. The excitation signal can be a 5-cycle Hanning window modulated sine pulse with a center frequency of 100kHz. This narrowband signal can effectively excite the guided waves of a specific mode while avoiding serious dispersion effects. By solving the wave equation, the propagation characteristics of ultrasonic guided waves in the pipeline can be obtained. For example, for the L(0,1) mode, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. The analysis of the dispersion curve is crucial to selecting a suitable excitation frequency. By comparing the dispersion curves of different modes, it can be found that the L(0,1) mode has a small dispersion in the low frequency band (such as 50-150kHz), which is suitable for actual detection. Selecting 100kHz as the excitation frequency can avoid serious dispersion effects while ensuring sufficient detection sensitivity. By calculating the displacement field and stress field distribution under different modes, the sensitivity of each mode to defects can be determined. For example, for the L(0,1) mode, the axial displacement is the largest in the middle of the pipe wall, which means that this mode is most sensitive to defects in the middle of the pipe wall. The tangential displacement of the T(0,1) mode is the largest on the surface of the pipe wall, so it is more sensitive to surface defects. The effect of temperature on the propagation characteristics of guided waves is an important research direction. By changing the temperature parameters in the model, the propagation behavior of guided waves at different temperatures can be simulated. For example, when the temperature increases from 20℃ to 100℃, the elastic modulus of the pipeline material may decrease by 5%-10%, which will lead to a decrease in the guided wave velocity.Specifically, the group velocity of the L(0,1) mode may decrease from 3000m / s to 2900m / s. At the same time, the increase in temperature will increase the acoustic impedance of the material, resulting in an increase in the attenuation of the amplitude of the waveguide signal. The least squares method is used to curve fit the waveguide propagation parameters at different temperatures, and the functional relationship between temperature and these parameters can be obtained. For example, there may be the following relationship between the waveguide velocity v and the temperature T: v = v0 (1-αT), where v0 is the velocity at the reference temperature and α is the temperature coefficient. By fitting the experimental data, it is possible to obtain the order of α≈10-4 / ℃. Using machine learning methods, such as support vector machines or neural networks, a mapping relationship between defect parameters and waveguide signal characteristics can be established. For example, a multi-layer perceptron model is trained using the arrival time, amplitude and frequency of the waveguide signal as input features and the type, location and size of the defect as output labels. By training on a large amount of simulated data, a model that can accurately predict defect parameters can be obtained. In practical applications, multiple sensors can be arranged around the pipeline to form a sensor array. By analyzing the signals received by different sensors, the location and size of defects can be estimated. For example, using time reversal technology, the position and shape of defects can be reconstructed through the reflected signals received by multiple sensors. This method has shown good application prospects in actual engineering and can effectively improve the accuracy and reliability of detection.

[0042] The differences between the first waveguide signal of the defect-free pipeline and the second waveguide signal of the defective pipeline are compared and analyzed, and the influence of defect type and size on the propagation time t, amplitude A and frequency f' of the waveguide signal are summarized.

[0043] According to the material properties and geometric dimensions of the multi-layer heterogeneous pipeline, a three-dimensional solid model of the pipeline is established, and appropriate boundary conditions and initial conditions are set according to the actual working conditions. The finite element method is used to mesh and discretize the pipeline model to obtain the finite element model of the pipeline. By solving the wave equation in the finite element model, the propagation process of ultrasonic guided waves in the pipeline is simulated, and the dispersion curves and modal vibration shapes at different frequencies are obtained. According to the dispersion curve, the propagation velocity and attenuation characteristics of the guided wave at different frequencies are analyzed, and the frequency range with stable propagation velocity and small attenuation is selected as the appropriate excitation frequency band. According to the modal vibration shape, the displacement field distribution characteristics of different guided wave modes are analyzed, and the guided wave mode with vibration shape energy concentrated on the inner wall of the pipeline and sensitive to defects is selected as the detection mode. Defect units of different types (such as cracks, corrosion, etc.) and sizes are introduced into the pipeline finite element model, and the defective pipeline is simulated by changing the material properties and geometric parameters of the defect unit. The same excitation load is applied to the finite element models of defect-free pipelines and defective pipelines respectively. The explicit dynamics finite element solver is used to calculate the guided wave propagation process within a certain time step, and the displacement time history of each node on the pipeline surface is obtained. The node displacement data is extracted as the guided wave signal. The guided wave signals of defect-free pipelines and defective pipelines are subjected to wavelet transform to extract the energy distribution characteristics of the guided wave signals in the time and frequency domains. The energy mutation of the guided wave signal at the defect position is focused on. By comparing and analyzing the differences in the characteristic parameters of the guided wave signal such as the arrival time of the defect reflection wave, the reflection wave amplitude and the frequency component, the mapping relationship between these characteristic parameters and the defect type and size is established. A defect characteristic parameter database is constructed, and the support vector machine classification algorithm is used to qualitatively identify the defect type, and the BP neural network regression algorithm is used to quantitatively evaluate the defect size. Through the training and testing of the machine learning algorithm, a defect qualitative and quantitative recognition model is obtained to realize the automatic recognition and evaluation of pipeline defects.

[0044] Specifically, the ultrasonic guided wave detection modeling of multi-layer heterogeneous pipelines first requires the establishment of an accurate three-dimensional solid model. Taking a three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. In finite element analysis, meshing is a key step. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, meeting the calculation accuracy requirements. In terms of boundary condition setting, axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. By solving the wave equation, the propagation characteristics of ultrasonic guided waves in the pipeline can be obtained. For example, for the L(0,1) mode, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. Choosing 100kHz as the excitation frequency can avoid severe dispersion effects while ensuring sufficient detection sensitivity. When introducing defects into the model, corrosion pits and cracks of different types and sizes can be considered. For example, a corrosion pit with a depth of 30% of the wall thickness and a width of 20mm, or a crack with a length of 10mm and a width of 0.5mm. By comparing the guided wave signals of defective and non-defective pipes, characteristic parameters such as the arrival time, amplitude and frequency of the reflected wave can be extracted. The energy distribution of the guided wave signal in the time-frequency domain can be analyzed using wavelet transform. For example, for the above-mentioned corrosion pit, a significant energy mutation may be observed at 200μs of the signal, while the crack may produce a weaker reflection at 180μs. These features can be used to distinguish different types of defects. After building a defect feature parameter database, machine learning methods can be used for defect identification. For example, the use of support vector machines can achieve classification of defect types with an accuracy of more than 95%. The use of BP neural networks can estimate the defect size with a relative error of less than 10%. This method can automate the defect identification process and greatly improve detection efficiency and accuracy.

[0045] A multi-layer heterogeneous pipeline ultrasonic guided wave detection experimental system was built, which included a signal generator, a power amplifier, a magnetostrictive transducer, a tested pipeline, an oscilloscope, and data acquisition and processing equipment.

[0046] According to the material properties and geometric dimensions of the pipeline, a three-dimensional solid model of the pipeline is established, including different material areas of the inner layer, outer layer and middle layer of the pipeline. The finite element analysis software ANSYS is used to mesh the pipeline model and generate a discretized finite element model of the pipeline. In the finite element model, the material property parameters of each layer of the pipeline are defined, including density, elastic modulus and Poisson's ratio. By solving the characteristic equation of the pipeline model, the phase velocity and group velocity data of the ultrasonic guided wave propagating in the pipeline at different frequencies are obtained. According to the obtained phase velocity and group velocity data, the dispersion curve of the ultrasonic guided wave propagating in the pipeline is drawn. The branch characteristics and cutoff frequency of the dispersion curve are analyzed to determine the appropriate excitation frequency range to avoid multi-modal superposition interference. The modal analysis tool in ANSYS is used to calculate the modal vibration shape and energy distribution of the ultrasonic guided wave at different frequencies. By observing the symmetry of the modal vibration shape and the concentration of the energy distribution, the propagation stability and attenuation rate of different modes are judged. Taking into account the single-mode frequency range of the dispersion curve, the symmetry of the modal vibration shape, and the uniformity of energy distribution, the optimal ultrasonic guided wave mode and frequency are selected as the detection parameters. The detection parameters are further optimized by simulating and analyzing the influence of defects of different types and sizes in the pipeline on the propagation characteristics of the selected mode. According to the optimized ultrasonic guided wave mode and frequency, the excitation signal parameters of the signal generator are set, including pulse width, repetition frequency, and amplitude. Select a suitable magnetostrictive transducer, and optimize the layout position and coupling mode of the transducer according to the size and material of the pipeline. The magnetostrictive transducer is driven by a power amplifier to excite ultrasonic guided waves of the required mode and frequency in the pipeline under test. Multiple receiving sensors are used to collect ultrasonic guided wave signals at different positions of the pipeline, and signal preprocessing and feature extraction are performed. By analyzing the changes in characteristic parameters such as amplitude, phase, and frequency of ultrasonic guided wave signals, the location and quantitative characterization of pipeline defects are achieved. Combined with the type, location, and size information of the defect, the health status and remaining life of the pipeline are evaluated. According to the detection results of the pipeline, suggestions for pipeline maintenance and replacement are given, and subsequent maintenance and inspection plans for the pipeline are formulated. Through regular ultrasonic guided wave testing, the health monitoring and management of pipelines throughout their life cycle can be achieved.

[0047] Specifically, establishing a three-dimensional solid model of the pipeline is the basis for ultrasonic guided wave detection. Taking a three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. In finite element analysis, meshing is a key step. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, meeting the calculation accuracy requirements. In terms of boundary condition setting, axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. By solving the wave equation, the propagation characteristics of ultrasonic guided waves in the pipeline can be obtained. For example, for the L(0,1) mode, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. Choosing 100kHz as the excitation frequency can avoid severe dispersion effects while ensuring sufficient detection sensitivity. When introducing defects into the model, corrosion pits and cracks of different types and sizes can be considered. For example, a corrosion pit with a depth of 30% of the wall thickness and a width of 20mm, or a crack with a length of 10mm and a width of 0.5mm. By comparing the guided wave signals of defective and non-defective pipes, characteristic parameters such as the arrival time, amplitude and frequency of the reflected wave can be extracted. The energy distribution of the guided wave signal in the time-frequency domain can be analyzed using wavelet transform. For example, for the above-mentioned corrosion pit, a significant energy mutation may be observed at 200μs of the signal, while the crack may produce a weaker reflection at 180μs. Selecting the right magnetostrictive transducer is crucial to exciting the required mode. For a pipe with a diameter of 200 mm, a ring transducer with a length of 100 mm can be selected, and a permanent magnet can be used to provide a bias magnetic field. The transducer is usually arranged about 100 mm from the end of the pipe to avoid interference from end reflections. The coupling method can be dry coupling or liquid coupling. The former is easy to operate, and the latter has better signal quality. In terms of signal processing, techniques such as wavelet denoising and Hilbert transform can be used to improve the signal-to-noise ratio. By analyzing the amplitude, phase and frequency characteristics of the signal, the location and quantitative characterization of the defect can be achieved. For example, the arrival time of the reflected signal can be used to determine the defect location, and the amplitude is related to the defect size. Combined with the defect characteristics, the health status and remaining life of the pipeline can be evaluated, providing a basis for maintenance decisions.

[0048] Adjust the position of the magnetostrictive transducer so that it can completely cover the pipe cross section.

[0049] According to the material properties and geometric dimensions of the pipeline, a three-dimensional solid model of the pipeline is established, including different material areas of the inner layer, outer layer and middle layer of the pipeline. Finite element analysis software such as ANSYS or COMSOL is used to mesh the pipeline model and set appropriate boundary conditions and load conditions. By solving the wave equation of the pipeline model, the dispersion curves and modal functions of ultrasonic guided waves under different modes are obtained. The dispersion curve describes the relationship between the phase velocity and group velocity of the guided wave and the frequency, which can be used to select a suitable excitation frequency range and avoid high-frequency dispersion areas. The modal function characterizes the displacement and stress distribution characteristics of different modal guided waves on the cross section of the pipeline, which can be used to determine which modes are more sensitive to defects of specific types and locations. Using the information of the dispersion curve and modal function, the excitation and reception scheme of ultrasonic guided waves is designed. The appropriate piezoelectric transducer array arrangement is selected, and the frequency, amplitude and phase of the excitation signal are optimized to achieve selective excitation and reception of the target modal guided wave. The energy attenuation law of different modal guided waves propagating in the pipeline is analyzed by simulation to determine the optimal amplification factor and propagation distance of the detection signal. According to the actual working conditions of the pipeline, such as changes in material properties caused by factors such as temperature and pressure, the material parameters in the mathematical model are modified accordingly. Through comparison and verification with experimental results, the mathematical model is continuously optimized and improved to improve the accuracy and reliability of simulation analysis. Using the optimized pipeline model, a large number of finite element simulation analyses are carried out to examine the influence of defects of different types, sizes and positions on the propagation characteristics of guided waves. The key parameters reflecting the characteristics of defects, such as the reflection coefficient, transmission coefficient, energy attenuation of guided waves, etc., are extracted to establish the corresponding relationship between the defect parameters and the characteristics of guided wave signals. Based on a large number of data samples obtained by simulation, machine learning algorithms such as support vector machine (SVM) or convolutional neural network (CNN) are used to train the defect quantitative identification model. Through feature engineering and parameter tuning, the generalization ability and robustness of the model are improved, and the automatic discrimination of the defect type and size of the actual detection data is realized. The trained defect recognition model is integrated into the ultrasonic guided wave detection system, and the collected measured signals are preprocessed, feature extracted and classified. Combined with the three-dimensional imaging results of the pipeline, the spatial location, type and severity of the defect are intuitively displayed, providing a reliable basis for pipeline safety assessment and maintenance decisions.

[0050] Specifically, establishing a three-dimensional solid model of a multi-layer heterogeneous pipeline is the basis for ultrasonic guided wave detection. Taking a three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. The elastic modulus, Poisson's ratio, density and other parameters of each layer of material need to be accurately set. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. In finite element analysis, meshing is a key step. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, meeting the calculation accuracy requirements. In terms of boundary condition setting, axial displacement excitation can be applied at one end of the pipeline to simulate the effect of the transducer. The outer surface is set as a free boundary, and the inner surface needs to consider the coupling effect with the fluid. By solving the wave equation, the propagation characteristics of ultrasonic guided waves in the pipeline can be obtained. For example, for the L(0,1) mode, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. Choosing 100kHz as the excitation frequency can avoid severe dispersion effects while ensuring sufficient detection sensitivity. When introducing defects into the model, corrosion pits and cracks of different types and sizes can be considered. For example, a corrosion pit with a depth of 30% of the wall thickness and a width of 20mm, or a crack with a length of 10mm and a width of 0.5mm. By comparing the guided wave signals of defective and non-defective pipes, characteristic parameters such as the arrival time, amplitude and frequency of the reflected wave can be extracted. The energy distribution of the guided wave signal in the time-frequency domain can be analyzed using wavelet transform. For example, for the above-mentioned corrosion pit, a significant energy mutation may be observed at 200μs of the signal, while the crack may produce a weaker reflection at 180μs. These features can be used to locate and identify defects. Choosing a suitable magnetostrictive transducer is crucial to excite the required mode. For a pipe with a diameter of 200 mm, a ring transducer with a length of 100 mm can be selected, and a permanent magnet is used to provide a bias magnetic field. The transducer is usually arranged about 100 mm away from the end of the pipe to avoid interference from end reflections. The coupling method can be dry coupling or liquid coupling. The former is easy to operate and the latter has better signal quality. In terms of signal processing, techniques such as wavelet denoising and Hilbert transform can be used to improve the signal-to-noise ratio. By analyzing the amplitude, phase and frequency characteristics of the signal, the location and quantitative characterization of the defect can be achieved. In order to improve the accuracy of defect identification, a machine learning algorithm can be used. For example, a support vector machine (SVM) is used to train a defect classification model. First, through a large number of simulations and experiments, the guided wave signal characteristics of defects of different types and sizes are obtained.Then, key features such as reflection coefficient and energy attenuation rate are extracted from these signals as inputs to the SVM. The classification performance of the model can be optimized by adjusting the kernel function and penalty parameters. Finally, the trained model is integrated into the detection system to realize automatic analysis of the measured signals. For example, the system can automatically identify a corrosion pit with a depth of 40% of the wall thickness and a length of 30mm, and give its precise location on the pipeline. This automated analysis greatly improves the detection efficiency and reliability, and provides strong support for pipeline safety assessment and maintenance decisions.

[0051] The excitation frequency f and the waveguide mode of the signal generator are set to excite and receive the third waveguide signal.

[0052] According to the material properties and geometric dimensions of the multi-layer heterogeneous pipeline, a mathematical model of the pipeline is established in finite element analysis software such as ANSYS, including different material property parameters of the inner layer, outer layer and middle layer of the pipeline, and meshing is performed. The dispersion characteristics of the established multi-layer heterogeneous pipeline model are analyzed using finite element analysis software to obtain the dispersion curves of the guided wave under different modes. Through the dispersion curve, the appropriate excitation frequency range is selected to avoid the frequency points where multi-mode superposition is serious. The vibration mode and energy distribution of the guided wave mode at different frequencies are obtained to screen out the dominant mode with long propagation distance and concentrated energy. The displacement field and stress field of the dominant mode are simulated and calculated to determine whether its propagation attenuation in the pipeline is less than the preset threshold to determine whether it meets the detection distance requirements. If the requirements are met, the mode is determined as the detection mode; if not, the dominant mode is re-screened until the requirements are met. According to the determined detection frequency and guided wave mode, the frequency, amplitude and other parameters of the excitation signal are set in the signal generator, and the excitation signal is generated and applied to the transducer. The transducer converts the electrical signal into mechanical vibration to excite the required waveguide mode. The waveguide signal propagating in the pipeline is collected by the receiving transducer, and the signal is conditioned by amplifiers, filters, etc., to finally obtain a waveguide signal that can be used for defect analysis. According to the changes in the amplitude, phase and other characteristics of the waveguide signal, it is possible to determine whether there is a defect in the pipeline and the location and size of the defect.

[0053] Specifically, finite element modeling of multi-layer heterogeneous pipelines is the basis of ultrasonic guided wave detection. Taking a three-layer composite pipeline as an example, the inner layer may be stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. When building a model in ANSYS, it is necessary to accurately set the elastic modulus, Poisson's ratio, density and other parameters of each layer of material. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. The difference in these parameters will significantly affect the propagation characteristics of ultrasonic waves. Meshing is a key step. For a pipeline with a diameter of 200mm and a wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, meeting the calculation accuracy requirements. Dispersion characteristic analysis is the key to selecting appropriate detection parameters. By solving the wave equation, the dispersion curves of different modal guided waves can be obtained. Taking the L(0,1) mode as an example, in the frequency range of 50-150kHz, the phase velocity may gradually increase from 3000m / s to 3500m / s, while the group velocity may decrease from 3200m / s to 3000m / s. This dispersion characteristic has an important impact on the propagation and interpretation of the signal. Selecting 100kHz as the excitation frequency can avoid serious dispersion effects while ensuring sufficient detection sensitivity. The selection of dominant modes requires consideration of multiple factors. By analyzing the vibration mode and energy distribution of the guided wave mode at different frequencies, the mode with long propagation distance and concentrated energy can be found. For example, for the above-mentioned pipeline, the L(0,2) mode may show good propagation characteristics near 100kHz, and its energy is mainly concentrated in the pipe wall, which is conducive to defect detection. By simulating and calculating the displacement field and stress field of this mode, if its attenuation within a propagation distance of 10m is less than 6dB, it can be considered to meet the detection requirements. The setting of the excitation signal directly affects the detection effect. According to the selected 100kHz frequency and L(0,2) mode, a 5-cycle Hanning window modulated sine wave with a peak voltage of 100 can be set in the signal generator.

[0054] V. This signal can not only effectively excite the target mode, but also control the bandwidth and reduce the influence of the dispersion effect. The selection and layout of the transducer are also crucial. For a pipe with a diameter of 200mm, a 100mm long annular magnetostrictive transducer can be selected, and a permanent magnet can be used to provide a bias magnetic field. The transducer is usually arranged about 100mm away from the end of the pipe to avoid interference from end reflection. The coupling method can be dry coupling or liquid coupling. The former is easy to operate and the latter has better signal quality. Signal acquisition and processing are the basis for the analysis of detection results. The received waveguide signal usually needs to be amplified by 20-40dB, and the low-frequency noise and high-frequency interference are removed by a bandpass filter. The passband of the filter can be set to 80-120kHz to match the frequency band of the excitation signal. The sampling rate should be at least 1MHz to ensure the complete reconstruction of the signal. Defect judgment and positioning are the ultimate goals of detection. By analyzing the characteristic changes of the amplitude, phase and other characteristics of the waveguide signal, it can be determined whether the pipeline has defects. For example, for a corrosion pit with a depth of 30% of the wall thickness and a width of 20 mm, an obvious reflected wave may be observed in the signal at a distance of 5 m from the transducer. The ratio of the amplitude of the reflected wave to the incident wave (reflection coefficient) can be used to estimate the severity of the defect. By measuring the arrival time of the reflected wave and the known group velocity, the defect position can be accurately located. In order to improve the reliability of detection, a multi-frequency and multi-modal comprehensive analysis method can be adopted. For example, in addition to the L (0, 2) mode, the T (0, 1) mode can also be used for detection at the same time. The two modes have different sensitivities to different types of defects, and comprehensive analysis can improve the accuracy and comprehensiveness of detection. In addition, by establishing a defect database and machine learning algorithms, automatic identification of defect types and sizes can be achieved, further improving detection efficiency and reliability.

[0055] An oscilloscope is used to observe the received third waveguide signal, and the third waveguide signal is transmitted to a data acquisition and processing device through a data acquisition card.

[0056] According to the material properties and geometric dimensions of multi-layer heterogeneous pipelines, the dispersion curve and modal characteristics of ultrasonic guided waves in pipelines are calculated using the finite element method. The appropriate ultrasonic guided wave mode and frequency range are selected to determine the parameters of the excitation signal. The magnetostrictive sensor is used to excite ultrasonic guided waves on the pipeline surface, and the received guided wave signal is observed using an oscilloscope. The received guided wave signal is digitized by a data acquisition card and transmitted to a computer. The digitized guided wave signal is short-time Fourier transformed using MATLAB to extract the amplitude and phase characteristics of the signal in the time and frequency domain. According to the theoretically calculated dispersion curve and modal characteristics, the mode decomposition method is used to determine the components of different modes in the received signal and their relative energy. Using the modal analysis results of the guided wave signal, combined with factors such as the reflection coefficient and propagation distance of the pipeline defect, the changes in each mode of guided wave signals caused by the defect are calculated. By comparing and analyzing the received signal with the reference signal in the defect-free state, the defects in the pipeline are identified and located. According to the change characteristics of the guided wave signal caused by the defect, the support vector machine classification algorithm is used to determine the type and size of the defect. Combining the location, type and size information of defects, a data-driven pipeline integrity assessment model is used to estimate the health status and remaining service life of the pipeline.

[0057] Specifically, ultrasonic guided wave detection of multi-layer heterogeneous pipelines is a complex engineering technology that requires the combination of knowledge from multiple disciplines such as materials science, acoustic principles, and signal processing. Take a typical three-layer composite pipeline as an example, the inner layer is stainless steel, the middle layer is polyethylene, and the outer layer is carbon steel. This structure is more common in the petrochemical industry and can take into account both corrosion resistance and mechanical strength. First, the dispersion curve and modal characteristics of ultrasonic guided waves are calculated using the finite element method. The key to this step is to accurately set material parameters and mesh division. For example, the elastic modulus of stainless steel is about 200GPa, polyethylene is 1GPa, and carbon steel is 210GPa. For a pipeline with a diameter of 200mm and a total wall thickness of 15mm, a hexahedral unit can be used, and the unit size is set to about 1mm. This ensures that at a frequency of 100kHz, each wavelength contains at least 20 units, meeting the calculation accuracy requirements. Through dispersion curve analysis, the appropriate guided wave mode and frequency range can be selected. For example, the L(0,2) mode may show good propagation characteristics in the range of 80-120kHz, with a group velocity of about 3200m / s and a phase velocity of about 3400m / s. The reason for choosing this mode and frequency range is that it can propagate over a long distance in the pipeline and is sensitive to defects. The design of the excitation signal directly affects the detection effect. A 5-cycle Hanning window modulated sine wave with a center frequency of 100kHz and a peak voltage of 100V can be selected. This signal can not only effectively excite the target mode, but also control the bandwidth and reduce the influence of the dispersion effect. The magnetostrictive sensor is an ideal ultrasonic guided wave excitation and receiving device. For a pipeline with a diameter of 200mm, a ring magnetostrictive transducer with a length of 100mm can be selected. Its working principle is to use the magnetostrictive effect to convert electromagnetic energy into mechanical energy, thereby generating ultrasonic waves on the surface of the pipeline. When receiving, it is a reverse process, converting mechanical vibration into electrical signals. The received guided wave signal usually needs to be amplified and filtered. For example, a 20-40dB amplifier can be used, and a bandpass filter can be used to remove low-frequency noise and high-frequency interference. The passband of the filter can be set to 80-120kHz to match the frequency band of the excitation signal. The sampling rate should be at least 1MHz to ensure complete reconstruction of the signal. Short-time Fourier transform (STFT) is an effective tool for analyzing the time-frequency characteristics of guided wave signals. Through STFT, the frequency components of the signal at different times can be obtained, thereby identifying guided waves of different modes. For example, if energy concentration is observed near 100kHz at a certain moment, and its propagation speed is consistent with the theoretically calculated L(0,2) mode, it can be determined that this is the signal of the target mode. The mode decomposition method can further separate the components of different modes. For example, for the above three-layer pipeline, in addition to the L(0,2) mode, modes such as L(0,1) and T(0,1) may also be excited. By comparing the relative energy of these modes, the excitation effect and signal quality can be evaluated. The core of defect detection is to analyze the changes in guided wave signals.For example, for a corrosion pit with a depth of 30% of the wall thickness and a width of 20 mm, a significant reflected wave may be observed in the signal at a distance of 5 m from the transducer. The ratio of the amplitude of the reflected wave to the incident wave (reflection coefficient) can be used to estimate the severity of the defect. By measuring the arrival time of the reflected wave and the known group velocity, the defect position can be accurately located. Support vector machine (SVM) is an effective defect classification algorithm. By establishing a defect database, the SVM model can be trained to identify defects of different types and sizes. For example, defects can be divided into three categories: corrosion, cracks, and pits, and classified according to the characteristics of the reflected signal (such as amplitude, frequency change, etc.). Finally, the data-driven pipeline integrity assessment model can comprehensively consider the location, type, and size of the defect to estimate the health status and remaining service life of the pipeline. For example, for the above corrosion pit, if its depth growth rate is 0.5 mm per year, it can be predicted that the wall thickness at this location will be reduced to below the safety threshold within 5 years, and it will need to be repaired or replaced within 4 years.

[0058] The collected third waveguide signal is subjected to time-frequency analysis to extract characteristic parameters of the propagation time t, amplitude A and frequency f' of the third waveguide signal.

[0059] According to the collected data of the third waveguide signal, the following processing steps are performed: the collected third waveguide signal is preprocessed, including using a wavelet denoising method to remove high-frequency noise in the signal, and using a bandpass filter to filter out low-frequency interference components in the signal, so as to obtain a preprocessed third waveguide signal. The preprocessed third waveguide signal is subjected to a short-time Fourier transform (STFT), and a suitable window type (such as a Hamming window), window length and overlap rate are set to obtain a time-frequency representation matrix of the signal. The time-frequency curve of the signal is extracted from the time-frequency representation matrix, and the curve reflects the frequency change of the signal at different times. The time-frequency curve is peak-detected, and the main frequency component of the signal and its corresponding time point are determined by setting an appropriate threshold. The selection of the main frequency component can be determined according to the peak amplitude or energy ratio. According to the time point of the main frequency component, the amplitude information at the corresponding time is extracted from the preprocessed third waveguide signal to obtain an amplitude-time curve. The amplitude-time curve is feature extracted to obtain the statistical characteristics of the curve, such as the maximum value, minimum value, average value, root mean square value, etc., as well as the shape characteristics of the curve, such as the peak-to-peak value, peak factor, pulse factor, etc. The frequency values, corresponding time points and amplitude characteristic parameters of the main frequency components are combined to form the characteristic vector of the third waveguide signal. In order to eliminate the differences in the dimensions and numerical ranges of different characteristic parameters, the characteristic vectors are normalized to the maximum and minimum values. The normalized characteristic vector is used as the comprehensive characteristic representation of the third waveguide signal for subsequent defect identification and classification tasks. According to the pre-established correspondence between the defect type and the characteristic vector, a machine learning algorithm such as support vector machine (SVM) or random forest is used to identify and classify the defect type of the newly collected third waveguide signal. Through the above steps, feature extraction and defect identification of the third waveguide signal can be achieved. In practical applications, it is also necessary to adjust and optimize the algorithm parameters according to the characteristics of the specific problem to obtain better recognition performance.

[0060] Specifically, preprocessing the third waveguide signal is a key step to improve the subsequent analysis effect. Wavelet denoising can effectively remove high-frequency noise. For example, using db4 wavelet, the decomposition layer number is 4, and the soft threshold method can remove more than 95% of high-frequency noise. Bandpass filtering can filter out low-frequency interference. For example, using a bandpass filter of 80-120kHz can retain the target frequency band signal. Short-time Fourier transform (STFT) can reveal the time-frequency characteristics of the signal. It is crucial to choose a suitable window function. For example, the Hamming window can achieve a good balance between time and frequency resolution. The window length can be set to 256 points with an overlap rate of 50%, which can not only ensure time-frequency resolution but also reduce the amount of calculation. Extracting the time-frequency curve from the STFT result can intuitively reflect the change of signal frequency over time. For example, for a pipeline with defects, a frequency mutation may be observed at a certain moment, which often corresponds to the defect location. By setting an amplitude threshold (such as 20% of the maximum amplitude) for peak detection, the main frequency component can be determined. According to the time point corresponding to the main frequency component, the amplitude information is extracted from the original signal to obtain the amplitude-time curve. This curve contains a wealth of defect feature information. For example, for crack defects, it may appear as a sudden drop in amplitude, while for corrosion defects, it may appear as a slow decay of amplitude. Extracting statistical features and shape features from the amplitude-time curve can fully describe the signal characteristics. Statistical features such as maximum and minimum values ​​reflect the overall intensity level of the signal, while shape features such as peak factor and pulse factor reflect the waveform characteristics of the signal. For example, crack defects may lead to a larger peak factor, while corrosion defects may appear as a smaller pulse factor. Combining frequency, time and amplitude features into feature vectors can fully characterize the third waveguide signal. By normalizing the maximum and minimum values, the dimensional differences between different features can be eliminated, improving the accuracy of subsequent classification. For example, all features are mapped to the [0, 1] interval so that different features have the same weight in classification. Based on the pre-established defect sample library, a machine learning algorithm is used for defect recognition and classification. Support vector machine (SVM) performs well in processing high-dimensional features and can effectively distinguish different types of defects. For example, using SVM with radial basis kernel function, it is possible to distinguish between cracks, corrosion, and dents with an accuracy of more than 95%. Random forest has good noise resistance and feature importance evaluation function, and can be used to screen the most discriminative features. Through the above steps, a comprehensive analysis of the third waveguide signal and defect identification can be achieved. This method can not only accurately identify the defect type, but also provide an estimate of the defect location and severity, providing an important basis for pipeline integrity assessment. In practical applications, it is necessary to optimize the algorithm parameters according to the specific pipeline material and structural characteristics to obtain the best detection effect.

[0061] A waveguide signal feature data set is constructed, which contains the characteristic parameters of waveguide signals under different defect types and sizes.

[0062] According to the material properties and structural parameters of the multi-layer heterogeneous pipeline, the geometric model and material model of the pipeline are established using finite element analysis software such as ANSYS, and appropriate boundary conditions and excitation conditions are set. The established pipeline model is meshed to obtain a discretized model of the pipeline. The dispersion curve and modal characteristics of ultrasonic guided waves at different frequencies are obtained by solving the characteristic equation of the pipeline discretized model. According to the dispersion curve, a suitable excitation frequency range is selected, and according to the modal characteristics, a suitable guided wave mode, such as L (0, 1) mode, is selected to obtain the optimal excitation parameters. In the pipeline geometric model, defects of different types and sizes, such as elliptical corrosion pits and rectangular cracks, are added by modeling, and the mesh of the defective area is refined to improve the simulation accuracy. The transient dynamic analysis module in ANSYS is used to perform time-domain finite element simulation calculations on the defective pipeline model to obtain guided wave propagation signals under different defect conditions. The simulated guided wave signal is processed by time-frequency analysis methods such as short-time Fourier transform (STFT) to extract parameters reflecting the defect characteristics, such as the amplitude, frequency and arrival time of the defect reflection wave. By changing the defect type, size and location, repeating steps 4-5, multiple sets of waveguide signal characteristic parameters under different defect conditions are obtained until the required defect type and size range are covered. The extracted waveguide signal characteristic parameters are classified and sorted according to the defect type and size, and a structured data set containing defect type, size and corresponding waveguide characteristic parameters is constructed for subsequent defect identification and quantitative analysis.

[0063] Specifically, the simulation of ultrasonic guided wave detection of multi-layer heterogeneous pipes is a complex engineering task that requires comprehensive consideration of material properties, structural parameters and defect characteristics. Take a common three-layer composite pipe as an example, the inner layer is polyethylene, the middle layer is steel, and the outer layer is epoxy resin coating. First, when establishing a geometric model in ANSYS, it is necessary to accurately define the thickness, density, elastic modulus and Poisson's ratio of each layer of material. For example, the thickness of the polyethylene layer is 2mm and the density is 0.95g / cm 3 , elastic modulus is 1.1GPa; steel layer thickness is 5mm, density is 7.85g / cm 3 , elastic modulus is 210GPa; epoxy resin layer thickness is 1mm, density is 1.2g / cm 3, the elastic modulus is 3.5GPa. In terms of boundary condition setting, free boundary conditions can be used to simulate the actual working state of the pipeline. The excitation condition can be set to apply axial displacement excitation at one end of the pipeline, with an amplitude of 1μm and a frequency range of 20-200kHz. Meshing is the key to simulation accuracy. For the above three-layer pipeline, at least 4 units can be divided into each layer in the radial direction, and the unit size in the axial direction does not exceed 1 / 10 of the minimum wavelength. This ensures the accuracy of the simulation results. The dispersion curve obtained by solving the characteristic equation can reveal the propagation characteristics of guided waves at different frequencies. For the above three-layer pipeline, it may be found that the L(0,1) mode has lower dispersion and higher excitation efficiency in the range of 50-100kHz. Therefore, 75kHz can be selected as the excitation frequency, and the L(0,1) mode can be used for defect detection. When adding defects, different types of typical defects can be considered. For example, simulate an elliptical corrosion pit with a depth of 50% of the steel layer thickness and a length of 10mm, or a rectangular crack with a width of 0.1mm and a depth of 30% of the steel layer thickness. The mesh of these defective areas is refined, and the unit size can be reduced to 1 / 4 of the surrounding area to improve the simulation accuracy. In transient dynamics analysis, the total simulation time can be set to 1ms and the time step can be set to 0.1μs to capture high-frequency components. When performing STFT analysis on the simulated signal, a 256-point Hamming window with an overlap rate of 50% can be selected. By analyzing the time-frequency diagram under different defect conditions, characteristic parameters such as the amplitude, main frequency and arrival time of the reflected wave can be extracted. By systematically changing the defect parameters, such as gradually increasing the corrosion pit depth from 10% to 90%, or changing the direction and position of the crack, a comprehensive data set can be established. This data set may contain hundreds of sets of guided wave characteristic parameters under different defect conditions, laying the foundation for subsequent defect identification and quantitative analysis. The advantage of this simulation method is that it can systematically study the effects of various defects on guided wave propagation without conducting a large number of physical tests. By analyzing the simulation results, the parameter selection in actual detection can be optimized to improve the accuracy and reliability of defect detection. At the same time, the data set generated by the simulation also provides valuable data resources for the training of machine learning algorithms, which helps to develop more intelligent and efficient defect identification systems.

[0064] The data set is divided into a training set and a test set, and the convolutional neural network model, BP neural network model and support vector machine model are trained respectively.

[0065] Preprocess the pipeline defect data, including noise removal and normalization, to improve data quality. The preprocessed data is used for subsequent feature extraction. According to the characteristics of pipeline defects, extract features that can characterize different types of defects, such as the shape, size, and location of the defects. The extracted features will be used to train the defect recognition model. Data enhancement technology is used to generate more training samples and expand the training data set by rotating, translating, scaling, and other transformations on the original data. The processed data set is divided into a training set and a test set in a ratio of 8:2. The training set is used for model training, and the test set is used to evaluate model performance. Construct a convolutional neural network model, design a network structure including convolutional layers, pooling layers, and fully connected layers, and use the Adam optimizer and cross entropy loss function to train the model. Construct a support vector machine model, select a Gaussian kernel function, and determine the optimal penalty coefficient C and kernel function parameter γ through grid search. Use the training set to train the model. Use the test set to evaluate the performance of the two models, and compare the defect recognition capabilities of the models by calculating indicators such as accuracy, precision, and recall. Among them, accuracy rate = number of correctly predicted samples / total number of samples, precision rate = true positive examples / (true positive examples + false positive examples), recall rate = true positive examples / (true positive examples + false negative examples). The model with the best performance is selected as the final defect recognition model. After preprocessing and feature extraction of the new pipeline defect data, it is input into the trained model for classification prediction and output of the defect type.

[0066] Specifically, preprocessing pipeline defect data is a key step to improve the accuracy of defect identification. First, wavelet transform can be used to remove high-frequency noise in the signal and retain effective defect features. For example, for a guided wave signal containing corrosion pits, using db4 wavelet for 5-layer decomposition can effectively filter out noise components above 50kHz. Then, the signal amplitude is mapped to the [0, 1] interval through maximum-minimum normalization to eliminate the amplitude difference under different detection conditions. Feature extraction is the core link of defect identification. For corrosion defects, features such as the peak amplitude, arrival time and frequency center of the reflected wave can be extracted. For example, a corrosion pit with a depth of 30% of the pipe wall thickness has a reflected wave amplitude of about 15% of the incident wave, and the arrival time is delayed by 0.5μs compared with the healthy pipeline. For crack defects, features such as the asymmetry of the waveform and the content of high-frequency components can be extracted. An axial crack will cause the L(0, 1) mode to be converted to the F(1, 2) mode, which appears as a new frequency component in the time-frequency diagram. Data enhancement technology can effectively expand the training samples. For pipeline defect data, new samples can be generated by adding random noise, time axis scaling, and amplitude modulation. For example, adding 5% Gaussian white noise to the original signal can simulate the detection results under different environments. Scaling the time axis by ±2% can simulate small changes in pipeline material parameters. These methods can expand the original data set by 3-5 times and improve the generalization ability of the model. When building a convolutional neural network model, a structure similar to VGG can be adopted. For example, 3 convolution blocks are used, each block contains 2 3x3 convolution layers and 1 maximum pooling layer, and finally 2 fully connected layers. Selecting ReLU as the activation function can speed up the convergence speed. The initial learning rate of the Adam optimizer is set to 0.001, and it decays by 10% every 50 epochs. To prevent overfitting, a Dropout layer can be added between the fully connected layers, and the dropout rate is set to 0.5. The core of the support vector machine model is the choice of kernel function. For pipeline defect recognition problems, the Gaussian kernel function usually performs better because it can capture nonlinear relationships in the feature space. When selecting the optimal parameters, the search range of the penalty coefficient C can be set to [0.1, 1, 10, 100], and the search range of the kernel function parameter γ can be set to [0.001, 0.01, 0.1, 1]. Through 5-fold cross validation, the optimal parameter combination that balances model complexity and generalization ability can be found. When evaluating model performance, in addition to the commonly used accuracy indicators, precision and recall should also be considered. For pipeline defect detection, the harm of missed detection (false negative examples) is usually greater than that of false detection (false positive examples). Therefore, more attention can be paid to the recall indicator. For example, for a test set containing 100 samples, 20 of which are defective samples, if the model correctly identifies 18 defects and misjudges 3 healthy samples as defects, the accuracy is 95%, the precision is 85.7%, and the recall is 90%. This shows that the model has good detection capabilities, but there is still room for improvement.The defect recognition model finally selected should be verified in actual applications. A new data set that was not involved in the training can be used for testing to evaluate the performance of the model under different working conditions. For example, a 10-meter-long pipeline is scanned, and a sensor is arranged every 1 meter to collect guided wave signals. After the signal is input into the model, if a corrosion pit with a depth of 20% at 3 meters in the pipeline can be accurately identified, and 6 meters and 9 meters can be correctly judged as healthy, then the model has good practicality.

[0067] The defect recognition accuracy of the three models was evaluated using the test set, and the model with the highest accuracy was selected as the final defect recognition model.

[0068] According to the defect signal data obtained from the pipeline defect detection experiment, the data is preprocessed. First, the wavelet threshold denoising method is used to reduce the noise of the signal to remove the high-frequency noise interference in the signal. Then, the denoised signal is normalized to the maximum and minimum values, and the signal amplitude is mapped to the [0, 1] interval to obtain the normalized signal data. The preprocessed defect signal is analyzed in time and frequency using continuous wavelet transform to extract the time and frequency domain characteristics of the signal. The Morlet wavelet is selected as the mother wavelet to perform multi-scale decomposition of the signal to obtain wavelet coefficients at different scales. According to the energy distribution of the wavelet coefficients, the time and frequency domain statistical characteristics of the signal are extracted, including the energy mean, variance, skewness and kurtosis at each scale, to form a feature vector. A convolutional neural network model is constructed for feature learning and classification recognition of defect signals. The network consists of three convolutional layers, three pooling layers and two fully connected layers. The convolution layer uses a 3×3 convolution kernel and the activation function is ReLU; the pooling layer uses a 2×2 maximum pooling operation; the number of neurons in the fully connected layer is 128 and 64 respectively, and the Dropout regularization method is used to prevent overfitting. The input of the network is the feature vector of the defect signal, and the output is the probability distribution of the defect type. A BP neural network model is constructed, which is also used for feature learning and classification recognition of defect signals. The network consists of an input layer, two hidden layers and an output layer. The number of neurons in the hidden layer is 64 and 32 respectively, and the activation function uses the Sigmoid function. The input and output of the network are the same as those of the convolutional neural network. The Gaussian kernel function is selected to construct a support vector machine model to classify the defect signal. The penalty coefficient C of the support vector machine and the parameter γ of the Gaussian kernel function are optimized by grid search and cross-validation methods to obtain the model parameters with the best performance. The defect recognition performance of the three models on the test set is evaluated using evaluation indicators such as accuracy, precision, recall and F1 score. By comparing the performance of different models, the model with the best comprehensive performance is selected as the final defect recognition model. On the basis of identifying the pipeline defect type, the defect location and defect size are further estimated. The defect area is segmented and extracted using the region growing algorithm, and the spatial position coordinates of the defect are calculated based on the segmentation results. The geometric size parameters of the defect, including length, width, and depth, are estimated through edge detection and contour fitting methods. The defect type, location, and size information are integrated to obtain a complete pipeline defect detection result, which is used to guide subsequent pipeline maintenance and decision-making.

[0069] Specifically, in pipeline defect detection, preprocessing is a key step to improve recognition accuracy. Wavelet threshold denoising can effectively filter out high-frequency noise and retain effective defect features. For example, for a guided wave signal containing corrosion pits, using db4 wavelet for 5-layer decomposition can remove noise components above 50kHz. Maximum and minimum value normalization maps the signal amplitude to the [0, 1] interval to eliminate the amplitude difference under different detection conditions. Time-frequency analysis is an important means to extract defect features. Continuous wavelet transform uses Morlet wavelet as the mother wavelet to clearly display the local features of the signal on the time-frequency plane. For an axial crack, its features are manifested as new frequency components on the time-frequency graph, such as the conversion of L(0, 1) mode to F(1, 2) mode. By calculating the statistical features of wavelet coefficients at different scales, such as energy mean, variance, etc., a feature vector reflecting the defect characteristics can be constructed. The convolutional neural network model performs well in the field of image recognition and is also suitable for feature learning and classification of defect signals. The network structure is similar to VGG, containing three convolution blocks, each with two 3×3 convolution layers and a maximum pooling layer. The use of the ReLU activation function can speed up convergence, and the Dropout layer helps prevent overfitting. For example, for a corrosion pit with a depth of 30% of the pipe wall thickness, a high recognition accuracy can be obtained after feature extraction and convolutional neural network processing. BP neural network is a classic machine learning model with a simple structure and is suitable for various classification problems. In defect recognition, two hidden layers can be used, with 64 and 32 neurons respectively. The Sigmoid activation function can map the output to the [0, 1] interval, which is easy to interpret as a probability. For crack-type defects, the BP neural network can effectively capture the asymmetric characteristics of the waveform. The core of the support vector machine model is the selection of the kernel function. The Gaussian kernel function can capture the nonlinear relationship in the feature space and is suitable for pipeline defect recognition problems. By optimizing the parameters through grid search, the best combination of balancing model complexity and generalization ability can be found. For example, the search range of the penalty coefficient C is set to [0.1, 1, 10, 100], and the range of the kernel function parameter γ is

[0070] [0.001, 0.01, 0.1, 1], the optimal parameters are selected by 5-fold cross validation. Model evaluation requires comprehensive consideration of multiple indicators. For pipeline defect detection, the harm of missed detection is usually greater than that of false detection, so more attention should be paid to recall rate. For example, in a test set containing 100 samples, if the model correctly identifies 18 defects (a total of 20) and misjudges 3 healthy samples as defects, the accuracy rate is 95%, the precision rate is 85.7%, and the recall rate is 90%. This shows that the model has good detection capabilities, but there is still room for improvement. Defect location and size estimation are important links in pipeline detection. The region growing algorithm can effectively segment the defect area and determine the defect location by calculating the centroid coordinates. Edge detection and contour fitting are used to estimate the geometric size of the defect. For example, a 10-meter-long pipeline is scanned and a sensor is arranged every 1 meter. If the model can accurately identify a corrosion pit with a depth of 20% at 3 meters and correctly judge other locations as healthy, it means that the model has good practicality. The fusion of this information provides an important basis for pipeline maintenance and decision-making.

[0071] Input the characteristic parameters of the waveguide signal of the pipeline to be tested into the trained defect recognition model, and output the recognition results of the defect type and size.

[0072] According to the dispersion curve and modal characteristics obtained by finite element simulation, the appropriate excitation frequency is selected so that the excited waveguide mode has a small dispersion effect and a high signal-to-noise ratio. According to the analysis results of the dispersion characteristics and waveguide mode, it is determined to use magnetostrictive sensors as excitation and receiving devices. According to the selected excitation frequency, the layout of the magnetostrictive sensor is designed to ensure that the sensor can effectively excite and receive the waveguide signal. The magnetostrictive sensor is used to excite ultrasonic guided waves on the surface of the pipeline and collect the waveguide signal propagating in the pipeline. The collected waveguide signal is preprocessed, including using wavelet transform for denoising and data standardization to improve the signal quality. The characteristic parameters of the time domain and frequency domain, such as amplitude, phase, frequency, etc., are extracted from the preprocessed waveguide signal as input features for defect identification. The extracted features are input into the pre-trained convolutional neural network model to identify common defect types such as cracks and corrosion in the pipeline. At the same time, the support vector regression model is used to quantitatively evaluate the size of the defect. The defect type recognition results of the convolutional neural network and the defect size assessment results of the support vector regression are combined to give the final pipeline defect detection results, including key information such as the defect type, location and size, providing a basis for pipeline safety assessment and maintenance decisions.

[0073] Specifically, the dispersion curve and modal characteristics analysis are the key to selecting a suitable excitation frequency. Taking steel pipes as an example, the dispersion curves of modes such as L(0, 1) and T(0, 1) can be obtained through finite element simulation. At a frequency of 100kHz, the group velocity of the L(0, 1) mode changes little, the dispersion effect is not obvious, and the wavelength is about 5cm, which is suitable for detecting centimeter-level defects. Therefore, choosing 100kHz as the excitation frequency can take into account both the dispersion effect and the detection accuracy. Magnetostrictive sensors have the advantages of non-contact and high efficiency, and are suitable for pipeline detection. For steel pipes with a diameter of 200mm, a circumferential magnetization method can be used, and a 20mm wide and 0.1mm thick nickel sheet can be used as the magnetostrictive material. Three turns of coils are evenly wound on the surface of the pipe, and ultrasonic guided waves are excited by pulse current. The receiving end uses a sensor with the same structure to effectively capture the reflected wave signal. Preprocessing of guided wave signals is crucial to improving detection accuracy. Using db4 wavelet for 5-layer decomposition can effectively remove high-frequency noise. For example, for a signal containing 50kHz noise, after wavelet denoising, the signal-to-noise ratio can be improved by more than 5dB. Maximum and minimum value normalization maps the signal amplitude to the [0, 1] interval to eliminate the amplitude difference under different detection conditions. Feature extraction is the basis of defect recognition. Time domain features include statistics such as signal peak, mean, variance, etc., which can reflect the overall characteristics of the defect. Frequency domain features such as spectrum energy distribution and main frequency can capture the frequency changes caused by defects. For example, for a corrosion pit with a depth of 30% of the pipe wall thickness, the energy proportion of its reflected signal in the 100-150kHz frequency band will increase significantly. Convolutional neural network models perform well in defect recognition. The network structure can include three convolution blocks, each with two 3×3 convolution layers and a maximum pooling layer. Using the ReLU activation function can speed up the convergence speed, and the Dropout layer helps prevent overfitting. For crack-type defects, the model can effectively capture the asymmetric characteristics of the waveform, and the recognition accuracy can reach more than 95%. The support vector regression model is suitable for quantitative evaluation of defect size. Selecting the Gaussian kernel function can capture the nonlinear relationship in the feature space. By optimizing the parameters through grid search, such as setting the search range of the penalty coefficient C to [0.1, 1, 10, 100] and the range of the kernel function parameter γ to [0.001, 0.01, 0.1, 1], the best parameter combination can be found. For the estimation of the depth of the corrosion pit, the average relative error of the model can be controlled within 10%. The comprehensive analysis results provide a basis for pipeline safety assessment. For example, a 10-meter-long pipeline is scanned and a sensor is arranged every 1 meter. If the model identifies a corrosion pit with a depth of 20% at 3 meters and correctly judges other locations as healthy, it means that the detection system has good practicality. This information can be used to formulate a reasonable maintenance plan, extend the service life of the pipeline, and reduce safety risks.

[0074] The above embodiments are only used to illustrate the technical solution of the present invention but not to limit it. The present invention is described in detail with reference to the preferred embodiments. It should be understood by those skilled in the art that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention, and should be included in the scope of the claims of the present invention.

Claims

1. A method for pipeline defect detection using the magnetostrictive characteristics of ultrasonic guided waves, characterized in that: include: Obtain the geometric and material parameters of multi-layer heterogeneous pipelines and establish a mathematical model of the pipeline; According to the mathematical model, the dispersion curve and modal characteristics of the ultrasonic guided wave in the pipeline are calculated; Select appropriate excitation frequency and waveguide mode to optimize ultrasonic guided wave detection parameters; Defects of different types and sizes are preset in the pipeline, and the influence of defects on guided wave propagation characteristics is analyzed through numerical simulation; Build an ultrasonic guided wave detection experimental system for multi-layer heterogeneous pipelines, collect guided wave signals under different defect conditions, and extract time-frequency domain characteristic parameters; Construct a waveguide signal feature data set, train convolutional neural network, BP neural network and support vector machine models, and obtain a defect quantitative recognition model; The guided wave signal of the pipeline to be tested is input into the trained defect recognition model, and the type, location and size information of the defect are output to achieve quantitative non-destructive detection and evaluation of pipeline defects.

2. The detection method according to claim 1, characterized in that The method of obtaining geometric and material parameters of a multi-layer heterogeneous pipeline includes: Measure the inner and outer diameters of the pipeline, thickness of each layer and other geometric parameters, and obtain material parameters such as density, elastic modulus and Poisson's ratio of each layer; The geometric and material parameters are dimensionless to eliminate the differences in dimensions and numerical ranges of different parameters.

3. The detection method according to claim 1, characterized in that The calculation of the dispersion curve and modal characteristics of ultrasonic guided waves in the pipeline includes: The semi-analytical finite element method and transfer matrix method are combined to solve the wave governing equation of the pipeline, and the phase velocity, group velocity, displacement and stress modes of the guided waves at different frequencies are obtained.

4. The detection method according to claim 1, characterized in that The detection parameters of the optimized ultrasonic guided wave include: Select the excitation frequency within the single-mode range according to the dispersion curve to avoid multi-mode superposition interference; According to the modal characteristics, the guided wave mode with low dispersion and high attenuation is selected to improve the detection sensitivity and signal-to-noise ratio.

5. The detection method according to claim 1, characterized in that The numerical simulation analysis of the influence of defects on the waveguide propagation characteristics includes: Defect units of different types, sizes and positions are introduced into the pipeline finite element model. The explicit dynamics algorithm is used to simulate the interaction between ultrasonic guided waves and defects. The amplitude, frequency and phase information of reflected and transmitted waves are extracted, and the mapping relationship between defect parameters and guided wave signal characteristics is established.

6. The detection method according to claim 1, characterized in that The experimental system for constructing a multi-layer heterogeneous pipeline ultrasonic guided wave detection system comprises: Use magnetostrictive sensors as excitation and receiving devices, matching appropriate excitation frequency and waveguide mode; The guided wave signals under different defect conditions are collected, and preprocessed such as denoising and normalization are performed to extract the statistical characteristics and energy distribution characteristics in the time and frequency domains.

7. The detection method according to claim 1, characterized in that The constructing of the waveguide signal feature data set comprises: Based on the waveguide signals collected from numerical simulation and experiments, characteristic parameters reflecting the defect type and size are extracted, and the characteristic data are labeled according to the defect parameters to form a structured data set for machine learning training.

8. The detection method according to claim 1, characterized in that The training of convolutional neural network, BP neural network and support vector machine models includes: Use data enhancement methods to expand training samples, optimize network structure and model parameters, and improve the accuracy, precision and recall of defect recognition; The cross-validation method is used to evaluate the model performance, and the model with the strongest generalization ability is selected as the final defect quantitative identification model.

Images