Pipeline ultrasonic multi-mode separation method and system and storage medium

Abstract

The invention discloses a pipeline ultrasonic multi-mode separation method and system and a storage medium, and the method comprises the steps: scanning a to-be-detected pipeline through a phased array ultrasonic detection system, and obtaining the full-matrix data of the to-be-detected pipeline; and inputting the full-matrix data of the to-be-detected pipeline into the trained auto-encoder for processing to obtain a defect echo signal of the to-be-detected pipeline. Wherein in the self-encoder training process, a training set is composed of full matrix data obtained according to different-mode ultrasonic waves and defect response models and different-mode ultrasonic waves and boundary response models, and random disturbance is further added; according to the method, training data of 12 different propagation modes are adopted for training during training, and the multi-mode fusion training strategy enables an auto-encoder to learn deep physical characteristics of ultrasonic propagation, so that the separation capability of defect echoes and boundary echoes is remarkably improved; and finally, clearer defect display and more excellent signal-to-noise ratio performance are realized in full-focusing imaging.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, and in particular to a method, system and storage medium for ultrasonic multimodal separation of pipelines. Background Technology

[0002] Ultrasonic phased array testing technology, as an advanced ultrasonic testing technique for pipelines, utilizes a computer to excite and control the reception of each transducer, achieving spatial deflection and focusing of the beam. However, when inspecting finite-sized structural components such as pipelines, the limited thickness of the ultrasonic waves causes multiple reflections and wave mode transitions within the specimen structure, typically generating numerous artifacts that severely impact defect detection and identification. Therefore, there is an urgent need to develop a wave packet separation method to isolate the inherent structural echo signals from the time-domain signal, effectively improving the imaging quality and signal-to-noise ratio of the imaging area.

[0003] Artifacts caused by the inherent echoes of the specimen structure during imaging are unavoidable when using full matrix capture (FMC) data. When using more complex ray paths to form different modes of ultrasonic imaging, these signals can lead to imaging artifacts at non-physical locations. In geophysics and ultrasonic nondestructive testing, the deconvolution method has been extensively studied for distinguishing overlapping signals; however, its performance in aluminum plate experiments is generally poor, especially when applied to single ultrasonic detection signals, and it exhibits weak noise immunity.

[0004] To overcome the problems of distortion, severe signal-to-noise ratio degradation, and poor stability associated with deconvolution methods, a convolutional neural network (CNN) is proposed for wave packet separation. The CNN learns the shape of overlapping signals in the time domain. It consists of multiple convolutional layers, adjusting the convolution kernel by learning the waveform. In the CNN model, some information is first encoded as an intermediate latent representation of the input data, reducing its dimensionality, and then this information is decoded into a structure similar to the input data. However, this method only uses a window function to zero out the intrinsic boundary echoes, failing to effectively separate the signal wave packets. Summary of the Invention

[0005] The purpose of this invention is to provide a method, system, and storage medium for ultrasonic multimodal separation of pipelines. The method utilizes a trained autoencoder to separate the defect echo signal of the pipeline under test from the full matrix data of the pipeline under test, which can effectively achieve the separation of signal packets and increase the subsequent imaging quality and signal-to-noise ratio.

[0006] To achieve the above objectives, the present invention employs the following technical solution:

[0007] In a first aspect, the present invention provides a method for ultrasonic multimodal separation of pipelines, comprising:

[0008] The pipeline under test is scanned using a phased array ultrasonic testing system to obtain full matrix data of the pipeline under test;

[0009] The full matrix data of the pipeline under test is input into a trained autoencoder for processing to obtain the defect echo signal of the pipeline under test.

[0010] Optionally, the step of obtaining the trained autoencoder includes:

[0011] Obtain the pre-set experimental parameters, including: phased array detection parameters, geometric condition parameters of defects and boundaries, and specimen parameters; the phased array detection parameters include: center frequency, number of array elements N, wedge angle α, and wedge height Z. s The distance between the front edge of the wedge and the edge; the geometric parameters of the defect and the boundary include: defect size, defect location, wedge interface, and bottom boundary of the specimen; the specimen parameters include: specimen thickness, specimen diameter, specimen density, transverse / longitudinal wave velocity of the specimen, weld shape, and weld size;

[0012] By simulating real detection conditions through random perturbation of grid points, some experimental parameters are gridded, and random perturbation is added to each grid point to obtain experimental parameters with perturbation.

[0013] The perturbation-containing experimental parameters are input into the constructed ultrasonic wave response models with different modes and the boundary response models with different modes to obtain full matrix data; the full matrix data includes ultrasonic wave information of different modes at the boundary and the defect, as well as the corresponding multiple echo information.

[0014] Preprocess the full matrix data to obtain the preprocessed full matrix data;

[0015] Construct a training set containing experimental parameters and preprocessed full matrix data, and use the defect locations in the experimental parameters as the predicted labels for the training set;

[0016] The training set is input into the autoencoder for training to obtain the defect locations detected in the specimen;

[0017] The autoencoder is trained by using the predicted labels from the training set and the defect locations detected in the specimen, and by minimizing the objective loss function to determine the optimal model parameters.

[0018] Optionally, the step of simulating real detection conditions by randomly perturbing grid points, gridding some experimental parameters, and adding random perturbations to each grid point to obtain perturbed experimental parameters includes:

[0019] S01: Determine the range of a parameter based on the value of a certain parameter among some experimental parameters; the certain parameters are specimen thickness, wedge angle, and defect size.

[0020] S02: Create a step size based on this parameter range. The structured mesh is used to obtain multiple mesh points;

[0021] S03: Based on step size And grid points, calculate the random perturbation for each grid point;

[0022] S04: Add a random perturbation at each grid point to obtain grid points with random perturbation;

[0023] Based on steps S01-S04, partial experimental parameters with random perturbations are obtained, and based on these partial experimental parameters with random perturbations, perturbation-containing experimental parameters are constructed.

[0024] Optionally, the formula for calculating the random perturbation is expressed as follows:

[0025] ;

[0026] in, Let m be the m-th grid point; For the random perturbation of the m-th grid point, This indicates the creation of two completely independent subsets from a network with random perturbations; The step size for the structured grid points.

[0027] Optionally, the preprocessing of the full matrix data to obtain preprocessed full matrix data includes:

[0028] The full matrix data is dimensionality reduced, and the dimensionality-reduced full matrix data is obtained.

[0029] The dimensionality-reduced full matrix data is linearly normalized so that the value range is kept within [0, 1], resulting in the preprocessed full matrix data.

[0030] Optionally, the different modes of ultrasonic waves and the defect response model are represented as follows:

[0031] ;

[0032] ;

[0033] ;

[0034] in, The excitation signal emitted by the i-th element; Indicates angular frequency; Indicates the location of the defect; The ultrasonic wave propagation modes are represented by LL, LT, TL, LT, TT, LL-L, LL-T, TT-L, TT-T, LT-L, LT-T, TL-L, and TL-T, where T represents transverse wave transmission, L represents longitudinal wave transmission, -T represents transverse wave reflection, and -L represents longitudinal wave reflection. The path time from the excitation of the i-th element to the reception of the j-th element; l is the imaginary unit; The path from the excitation of the i-th element to the reception of the j-th element. The scattering coefficient at that location; , These are the transmission coefficients of the excitation path of the i-th element and the receiving path of the j-th element through different media layers, respectively. , These are the reflection coefficients of the excitation path of the i-th element and the receiving path of the j-th element through different media layers, respectively. , Let be the diffusion attenuation coefficients of ultrasonic wave propagation on the excitation path of the i-th element and the receiving path of the j-th element, respectively. Let be the material attenuation coefficient for ultrasonic wave propagation along the excitation path of the i-th element and the receiving path of the j-th element. , These represent the far-field directivity of the i-th element excitation path and the j-th element receiving path under different ultrasonic wave propagation modes. For the i-th element and The included angle at the point; and These are the propagation processes of the excitation of the i-th element and the reception of the j-th element, respectively. The formula and Same format;

[0035] in, ;

[0036] ;

[0037] in, The interface position along the k-path of ultrasonic wave propagation; Let be the angle of incidence at the interface along the k-path of ultrasonic wave propagation; The angle of refraction or reflection at the interface along the k-path of ultrasonic wave propagation; Let k be the speed of ultrasonic wave propagation along path k. Let n be the refractive index at the interface along the ultrasonic wave propagation path k; k is the ultrasonic wave propagation path, and n is the number of transmissions. Different incident waves, including incident transverse waves and incident longitudinal waves;

[0038] in, ;

[0039] ;

[0040] ;

[0041] ;

[0042] ;

[0043] ;

[0044] ;

[0045] Wherein, LL and LT represent longitudinal wave transmission and transverse wave transmission under longitudinal wave incident conditions, respectively; TL and TT represent longitudinal wave transmission and transverse wave transmission under transverse wave incident conditions, respectively. The transmission coefficients of longitudinal and transverse waves under longitudinal wave incidence conditions; The transmission coefficients of longitudinal and transverse waves under transverse wave incidence conditions; These are the angles between the incident longitudinal wave and the normal, and the angles between the incident transverse wave and the normal, respectively. These are the refraction angles of the longitudinal and transverse waves, respectively, under the condition of longitudinal wave incidence. These are the refraction angles of the longitudinal and transverse waves under the condition of transverse wave incidence, respectively; These are the reflection angles of the transverse wave under longitudinal wave incidence and transverse wave incidence conditions, respectively. These are the material densities of the specimen and the wedge block, respectively. These are the longitudinal wave velocity of the specimen, the transverse wave velocity of the specimen, the longitudinal wave velocity of the wedge, and the transverse wave velocity of the wedge, respectively.

[0046] in, ;

[0047] ;

[0048] ;

[0049] ;

[0050] ;

[0051] ;

[0052] in, These represent longitudinal wave reflection and transverse wave reflection under longitudinal wave incidence conditions, respectively. These represent longitudinal wave reflection and transverse wave reflection under transverse wave incidence conditions, respectively. The reflection coefficients of longitudinal and transverse waves under longitudinal wave incidence conditions; The reflection coefficients of longitudinal and transverse waves under transverse wave incidence conditions; These are the angles between the sensor normal and the incident longitudinal wave, and the angles between the sensor normal and the incident transverse wave, respectively. These are the reflection angles of the transverse wave under longitudinal wave incidence and transverse wave incidence conditions, respectively. These are the longitudinal wave velocity and the transverse wave velocity, respectively.

[0053] Optionally, the different mode ultrasonic waves and boundary response models are defined by comparing the different mode ultrasonic waves with the defect response models representing the defect locations. Replace with a vector representing the position of the ultrasonic reflection point on the boundary. constitute.

[0054] Optionally, the autoencoder includes an encoder and a decoder connected in sequence; the encoder includes eight three-dimensional convolutional blocks connected in sequence; the decoder includes one three-dimensional convolutional block.

[0055] Secondly, the present invention provides a fault diagnosis system for acoustic and vibration data fusion of rotating equipment, comprising:

[0056] The data acquisition module is used to scan the pipeline under test using a phased array ultrasonic testing system to obtain the full matrix data of the pipeline under test.

[0057] The data processing module is used to input the full matrix data of the pipeline under test into the trained autoencoder for processing, and obtain the defect echo signal of the pipeline under test.

[0058] Thirdly, the present invention provides a computer-readable storage medium storing a computer program that, when executed, implements the pipeline ultrasonic multimodal separation method described in the first aspect.

[0059] Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

[0060] This invention provides a method, system, and storage medium for multimodal ultrasonic separation of pipelines. The method utilizes a trained autoencoder to process the full matrix data of the pipeline under test to separate the defect echo signal. During the training process of the autoencoder, the training set is composed of full matrix data obtained from different modes of ultrasonic wave and defect response models and different modes of ultrasonic wave and boundary response models, and random perturbation is also added. Training is conducted using training data from 12 different propagation modes. This multimodal fusion training strategy enables the autoencoder to learn the deep physical characteristics of ultrasonic wave propagation, thereby significantly improving the ability to separate defect echoes from boundary echoes. Ultimately, this results in clearer defect display and better signal-to-noise ratio performance in full-focus imaging.

[0061] This invention provides a method, system, and storage medium for multimodal separation of ultrasonic waves in pipelines. The method employs an autoencoder structure to directly learn and separate multimodal ultrasonic packets from time-domain signals, overcoming the shortcomings of traditional frequency-domain deconvolution methods, such as poor noise resistance and bandwidth dependence, and significantly improving the signal-to-noise ratio and imaging clarity.

[0062] This invention provides a method, system, and storage medium for multimodal separation of ultrasonic waves in pipelines. The method integrates a forward model (two response models) and introduces random noise and parameter perturbations to enhance the adaptability of the autoencoder to complex acoustic paths and noise in actual detection. Attached Figure Description

[0063] Figure 1 The diagram shown is a schematic flowchart of a pipeline ultrasonic multimodal separation method in one embodiment of the present invention.

[0064] Figure 2 The diagram shown is a schematic diagram of the training process of the autoencoder in one embodiment of the present invention.

[0065] Figure 3 The diagram shown is a schematic representation of the ultrasonic propagation path under longitudinal wave incident conditions in one embodiment of the present invention.

[0066] Figure 4 The diagram shown is a schematic representation of the ultrasonic propagation path under transverse wave incident conditions in one embodiment of the present invention.

[0067] Figure 5 The diagram shown is a schematic diagram of the self-encoder model structure in one embodiment of the present invention;

[0068] Figure 6 The figure shown is a training performance curve of the autoencoder in one embodiment of the present invention.

[0069] Figure 7 The image shown is a diagram of the received signal of a 12-element array in a small-diameter pipe weld inspection experiment according to one embodiment of the present invention.

[0070] Figure 8 The diagram shown is a picture of the received signals of all array elements in a small-diameter pipe weld inspection experiment according to one embodiment of the present invention.

[0071] Figure 9 The image shown is an ultrasonic full-focusing imaging diagram of pore defects in a small-diameter pipe in one embodiment of the present invention.

[0072] Figure 10 The image shown is a full-focus imaging result of different modes after wave packet separation in one embodiment of the present invention. Detailed Implementation

[0073] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0074] Example 1

[0075] like Figure 1 As shown in the figure, this invention provides a method for ultrasonic multimodal separation in pipelines, comprising the following steps:

[0076] S01: Use a phased array ultrasonic testing system to scan the pipeline under test to obtain the full matrix data of the pipeline under test;

[0077] S02: Input the full matrix data of the pipeline under test into the trained autoencoder for processing to obtain the defect echo signal of the pipeline under test.

[0078] Specifically, unlike traditional ultrasonic probes which have only one chip, phased array ultrasonic testing systems consist of multiple independent chips (typically 16, 32, 64, or 128). Each chip can be independently controlled by an electronic system to emit and receive ultrasonic waves individually or in a specific timing sequence.

[0079] Specifically, such as Figure 2 As shown, the training process of the autoencoder in step S02 includes:

[0080] S21: Obtain the set experimental parameters, including: phase detection parameters, geometric condition parameters of defects and boundaries, and specimen parameters;

[0081] Specifically, the phased array detection parameters include: center frequency, number of array elements N, wedge angle α, and wedge height Z. s The distance between the front edge of the wedge and the edge; the geometric parameters of the defect and the boundary include: defect size, defect location, wedge interface, and bottom boundary of the specimen; the specimen parameters include: specimen thickness, specimen diameter, specimen density, transverse / longitudinal wave velocity of the specimen, weld shape, and weld size;

[0082] The center frequency is the main frequency of the ultrasonic pulses emitted by the probe; the wedge front distance is the horizontal distance from the foremost element of the probe to the front end of the wedge. This parameter is crucial for accurate defect location, especially when detecting near-surface defects or performing dimensional measurements.

[0083] S22: Simulate real detection conditions by randomly perturbing grid points, grid some experimental parameters, and add random perturbations to each grid point to obtain perturbed experimental parameters.

[0084] S23: Input the perturbation-containing experimental parameters into the constructed ultrasonic wave and defect response models and ultrasonic wave and boundary response models of different modes to obtain the full matrix data;

[0085] The full matrix data includes ultrasonic information of different modes at the boundary and the defect, as well as the corresponding multiple echo information;

[0086] S24: Preprocess the full matrix data to obtain the preprocessed full matrix data;

[0087] S25: Construct a training set containing experimental parameters and preprocessed full matrix data, and use the defect locations in the experimental parameters as the predicted labels of the training set;

[0088] S26: Input the training set into the autoencoder for training to obtain the defect location detected in the specimen;

[0089] S27: Based on the predicted labels of the training set and the defect locations detected in the specimen, the optimal model parameters are determined by minimizing the target loss function, thereby obtaining the trained autoencoder.

[0090] Specifically, in step S22, some experimental parameters are gridded by simulating real detection conditions through random perturbation of grid points, and random perturbation is added to each grid point to obtain experimental parameters with perturbation, including:

[0091] S01: Determine the range of a parameter based on the value of a certain parameter among some experimental parameters; the certain parameters are specimen thickness, wedge angle, and defect size.

[0092] S02: Create a step size based on this parameter range. The structured mesh is used to obtain multiple mesh points;

[0093] S03: Based on step size And grid points, calculate the random perturbation for each grid point;

[0094] Specifically, the formula for calculating random perturbations is expressed as follows:

[0095] ;

[0096] in, Let m be the m-th grid point; For the random perturbation of the m-th grid point, This indicates the creation of two completely independent subsets from a network with random perturbations; The step size for the structured grid points.

[0097] S04: Add a random perturbation at each grid point to obtain grid points with random perturbation;

[0098] Specifically,

[0099] in, Let m be the m-th grid point with random perturbation.

[0100] Based on steps S01-S04, partial experimental parameters with random perturbations are obtained, and based on these partial experimental parameters with random perturbations, perturbation-containing experimental parameters are constructed.

[0101] Specifically, random perturbations are added to the original experimental parameters, which distort the signal waveform, generate noise interference, and reduce the signal-to-noise ratio, thus simulating actual signal detection. This prevents idealized simulated signals from being unsuitable for real-world applications.

[0102] In this embodiment, the transmitted longitudinal wave (L), transmitted transverse wave (T), and inner wall reflected wave are considered; from Figure 3 and Figure 4 It can be seen that there are 4+8 propagation modes of ultrasound. The propagation of ultrasound direct wave signals can be divided into 4 propagation modes: LL, LT, TL (which is reciprocal with LT), and TT, and 8 propagation modes including one inner wall echo: LL-L, LL-T, TT-L, TT-T, LT-L, LT-T, TL-L, and TL-T.

[0103] Figure 3 In the diagram, dashed lines represent transverse waves, and solid lines represent longitudinal waves; The sensor normal direction and the incident longitudinal wave to Angle at the interface; For incident longitudinal wave and The angle between interface normals; For the reflected transverse wave in the wedge and The angle between interface normals; + This is the angle generated by the LT mode; For the reflected longitudinal wave in the wedge and The angle between interface normals; + This is the angle generated by the LL mode; For the refracted transverse wave in the specimen and The angle between interface normals; for At the interface, reflected transverse waves and The angle between interface normals; for At the interface, reflected longitudinal waves and The angle between the interface normals; under longitudinal wave incidence, + This refers to the included angle generated in the specimen under LT-L mode; under longitudinal wave incidence, + This refers to the included angle generated in the specimen under the LT-T mode; For the refracted longitudinal wave in the specimen and The angle between interface normals; for At the interface, reflected transverse waves and The angle between interface normals; for At the interface, reflected longitudinal waves and The angle between the interface normals; under longitudinal wave incidence, + This refers to the included angle generated in the specimen under LL-T mode; under longitudinal wave incidence, + This refers to the included angle generated in the specimen under the LL-L mode;

[0104] Figure 4 middle, The sensor normal direction and the incident transverse wave to Angle at the interface; For incident transverse waves and The angle between interface normals; For the reflected transverse wave in the wedge and The angle between interface normals; + This is the angle generated by the TT mode; For the reflected longitudinal wave in the wedge and The angle between interface normals; + This is the angle generated by the TL mode; For the refracted transverse wave in the specimen and The angle between interface normals; for At the interface, reflected transverse waves and The angle between interface normals; for At the interface, reflected longitudinal waves and The angle between the interface normals; under transverse wave incidence, + This refers to the included angle generated in the specimen under TT-T mode; under transverse wave incidence, + This refers to the included angle generated in the specimen under the TT-L mode; For the refracted longitudinal wave in the specimen and The angle between interface normals; for At the interface, reflected transverse waves and The angle between interface normals; for At the interface, reflected longitudinal waves and The angle between the interface normals; under transverse wave incidence, + This refers to the included angle generated in the specimen under TL-T mode; under transverse wave incidence, + This refers to the included angle generated in the specimen under the TL-L mode.

[0105] Based on the above understanding of the 12 modes, ultrasonic wave response models and boundary response models for different modes are established to obtain full matrix data.

[0106] The different ultrasonic modes and defect response models are represented as follows:

[0107] ;

[0108] ;

[0109] ;

[0110] in, The excitation signal emitted by the i-th element; Indicates angular frequency; Indicates the location of the defect; The ultrasonic wave propagation modes are represented by LL, LT, TL, LT, TT, LL-L, LL-T, TT-L, TT-T, LT-L, LT-T, TL-L, and TL-T, where T represents transverse wave transmission, L represents longitudinal wave transmission, -T represents transverse wave reflection, and -L represents longitudinal wave reflection. The path time from the excitation of the i-th element to the reception of the j-th element; l is the imaginary unit; The path from the excitation of the i-th element to the reception of the j-th element. The scattering coefficient at that location; , These are the transmission coefficients of the excitation path of the i-th element and the receiving path of the j-th element through different media layers, respectively. , These are the reflection coefficients of the excitation path of the i-th element and the receiving path of the j-th element through different media layers, respectively. , Let be the diffusion attenuation coefficients of ultrasonic wave propagation on the excitation path of the i-th element and the receiving path of the j-th element, respectively. Let be the material attenuation coefficient for ultrasonic wave propagation along the excitation path of the i-th element and the receiving path of the j-th element. , These are the far-field directivity of the i-th element excitation path and the j-th element receiving path in different ultrasonic propagation modes, respectively, i.e. the characteristics of the sound pressure intensity emitted or received by the array element changing with spatial direction, and changing with angle and angular frequency. For the i-th element and The included angle at the point; and These are the propagation processes of the excitation of the i-th element and the reception of the j-th element, respectively.

[0111] Consider the beam spread of the ultrasonic wave after multiple reflections on the upper and lower surfaces of the specimen. It can be seen that after n transmissions and reflections (n-1 reflections), the ultrasonic signal attenuates due to diffusion as follows:

[0112] ;

[0113] ;

[0114] in, The interface position along the k-path of ultrasonic wave propagation; Let be the angle of incidence at the interface along the k-path of ultrasonic wave propagation; The angle of refraction or reflection at the interface along the k-path of ultrasonic wave propagation; Let k be the speed of ultrasonic wave propagation along path k. Let n be the refractive index at the interface along the ultrasonic wave propagation path k; k is the ultrasonic wave propagation path, and n is the number of transmissions. For different incident waves, including incident transverse waves and incident longitudinal waves; p is the sequence number;

[0115] in, The calculation formula and The format is the same, so I will not repeat it here;

[0116] in, ;

[0117] ;

[0118] ;

[0119] ;

[0120] ;

[0121] ;

[0122] ;

[0123] Wherein, LL and LT represent longitudinal wave transmission and transverse wave transmission under longitudinal wave incident conditions, respectively; TL and TT represent longitudinal wave transmission and transverse wave transmission under transverse wave incident conditions, respectively. The transmission coefficients of longitudinal and transverse waves under longitudinal wave incidence conditions; The transmission coefficients of longitudinal and transverse waves under transverse wave incidence conditions; These are the angles between the incident longitudinal wave and the normal, and the angles between the incident transverse wave and the normal, respectively. These are the refraction angles of the longitudinal and transverse waves, respectively, under the condition of longitudinal wave incidence. These are the refraction angles of the longitudinal and transverse waves under the condition of transverse wave incidence, respectively; These are the reflection angles of the transverse wave under longitudinal wave incidence and transverse wave incidence conditions, respectively. These are the material densities of the specimen and the wedge block, respectively. These are the longitudinal wave velocity of the specimen, the transverse wave velocity of the specimen, the longitudinal wave velocity of the wedge, and the transverse wave velocity of the wedge, respectively.

[0124] in, The calculation formula and The format is the same, so I will not repeat it here;

[0125] in, ;

[0126] ;

[0127] ;

[0128] ;

[0129] ;

[0130] ;

[0131] in, These represent longitudinal wave reflection and transverse wave reflection under longitudinal wave incidence conditions, respectively. These represent longitudinal wave reflection and transverse wave reflection under transverse wave incidence conditions, respectively. The reflection coefficients of longitudinal and transverse waves under longitudinal wave incidence conditions; The reflection coefficients of longitudinal and transverse waves under transverse wave incidence conditions; These are the angles between the sensor normal and the incident longitudinal wave, and the angles between the sensor normal and the incident transverse wave, respectively. These are the reflection angles of the transverse wave under longitudinal wave incidence and transverse wave incidence conditions, respectively. These are the longitudinal wave velocity and the transverse wave velocity, respectively.

[0132] in, The calculation formula and The format is the same, so I won't repeat it here.

[0133] Based on the above-mentioned ultrasonic wave and defect response models of different modes, ultrasonic wave and boundary response models of different modes are constructed, including multiple echoes at the wedge interface during the detection process and ultrasonic wave information of different modes at the boundary of the specimen itself.

[0134] Different mode ultrasonic boundary response models do not need to consider scattering by scatterers in the specimen. The calculated propagation process can be divided into two parts: the excitation process reaching the interface or bottom surface and the reception process reflected by the interface or bottom surface. The different mode ultrasonic waves and boundary response models are used by integrating the different mode ultrasonic waves with the defect response model representing the defect location. Replace with a vector representing the position of the ultrasonic reflection point on the boundary. constitute.

[0135] Specifically, in this embodiment, the preprocessing step of the full matrix data in S24 includes:

[0136] The full matrix data is dimensionality reduced, and the dimensionality-reduced full matrix data is obtained.

[0137] The dimensionality-reduced full matrix data is linearly normalized so that the value range is kept within [0, 1], resulting in the preprocessed full matrix data.

[0138] Specifically, in this embodiment, in S25, a training set containing experimental parameters and preprocessed full matrix data is constructed, and the defect location in the experimental parameters is used as the prediction label of the training set. When constructing the training set, since both the defect echo and the specimen boundary echo exhibit a smooth shifting relationship across the full matrix data, less array information is used to effectively represent the signal characteristics of the full matrix data. Eight independent excitation and receiving elements are selected at equal intervals within the complete array of an ultrasonic phased array linear probe (1×32). This is equivalent to using a phased array probe with the same aperture size as the original, but with a reduced number of elements and increased spacing. The extracted signal... This is a subset of the original full matrix data. The construction relation of the dataset can be represented as a mapping function of the selected subset. Where R (32×32×2480) is the actual full matrix data collected in the experiment. This embodiment uses less array information to effectively represent the signal characteristics of the full matrix data, thereby reducing computational complexity.

[0139] In this embodiment, the self-encoder includes an encoder and a decoder connected in sequence, such as Figure 5As shown, the encoder comprises eight sequentially connected 3D convolutional blocks; the decoder comprises one 3D convolutional block. The model structure is based on a volumetric segmentation architecture to extract 3D convolutional layer information from the temporal envelope signal. To make the neural network model more accurate and robust, the input includes not only the detection signal generated by the forward model (two response models) but also Gaussian noise to adapt to changes in real experimental conditions. Linear activation functions are used throughout the inner layers and the output layer, with a dense layer reflecting defect information.

[0140] Specifically, the autoencoder employs the ADAM optimizer and is independently trained based on a scheduled learning rate. The workflow is as follows: first, the temporal trajectory of the signal is encoded; second, the feature parameters are decoded to obtain the defect location. The output data requires calling the array detection model to calculate different pattern signals at the corresponding locations and then outputting them.

[0141] Example 2

[0142] Based on the ultrasonic multimodal separation method for pipelines provided in Embodiment 1, this embodiment further illustrates the present invention in conjunction with specific embodiments.

[0143] This embodiment utilizes an ultrasonic phased array detection system to take the detection of weld seams in austenitic stainless steel small-diameter pipes containing pores as an example, and verifies in detail the effectiveness of the ultrasonic multimodal separation method for pipelines in this embodiment.

[0144] 1) Construct an ultrasonic testing experimental system;

[0145] A linear phased array probe with a center frequency of 5MHz and N=32 elements was used. The material being tested was a small-diameter austenitic stainless steel tube with dimensions of φ51mm×6mm×300mm. The weld zone was trapezoidal, with a weld width of 10mm at the top and 4mm at the root. An angled... A wedge with a height of 20° and Zs = 9.35 mm (from the center of the array to the surface of the specimen) is coupled to a small-diameter pipe. The density and longitudinal wave velocity of the wedge are respectively... =1000 kg / m 3 and =2670 m / s.

[0146] 2) Construct a training dataset for small-diameter austenitic stainless steel pipes;

[0147] Training data was calculated every 0.2 mm within a range of 0.2 mm to 8.4 mm from the leading edge of the wedge. Defect echo data at different horizontal positions were classified into labels 1 to 42, and defect-free data were categorized into label 43. In addition, random parameters were added under each label. These parameters varied in the following parameter space: thickness ∈[5.8mm, 6.2mm], wedge angle The radius of the circular hole is r ∈ [18.5°, 20.5°], and the radius of the circular hole is r ∈ [1.45mm, 1.55mm]. Each label in this dataset has 120 data sets, totaling 5160 data sets, of which 580 sets are used for testing, 580 sets for validation, and 4000 sets for training. The data on pore defects in the pipe are trained using 8 elements on the side closest to the weld.

[0148] 4) Input the training dataset into the autoencoder, with a training period of 30 and a learning rate of 1×10⁻⁶. -4 The regularization parameter is 1×10. -4 After training, the defect identification accuracy stabilized at 99.78%, as... Figure 6 As shown, no overfitting occurred.

[0149] 5) Ultrasonic phased array detection experiments were conducted on the porosity defects in austenitic stainless steel small-diameter pipe specimens, and full matrix data of the austenitic stainless steel small-diameter pipe specimens were collected; such as... Figure 7 and Figure 8 The diagram shows the received signal of 12 array elements and the received signal of all array elements in the small-diameter pipe weld inspection experiment.

[0150] 6) Image processing of the full matrix data of austenitic stainless steel small-diameter pipe specimens, such as... Figure 9 The multi-mode full-focus imaging results of the small-diameter tube experiment shown can be seen to show that there are significant differences in the imaging results of different modes, and the signal-to-noise ratio is generally low.

[0151] 7) The full matrix data of the austenitic stainless steel small-diameter pipe specimen was input into the autoencoder for multimodal separation, and the imaging results are as follows: Figure 9 As shown.

[0152] In summary, the imaging results demonstrate that this method can separate wave packets of defect echoes of different modes with limited input data, and can be used for ultrasonic wave packet separation of pipelines to achieve accurate imaging.

[0153] The above are typical applications of the present invention, but the applications of the present invention are not limited thereto.

[0154] Example 3

[0155] This invention describes a system for implementing the ultrasonic multimodal separation method for pipelines provided in Embodiment 1, comprising:

[0156] The data acquisition module is used to scan the pipeline under test using a phased array ultrasonic testing system to obtain the full matrix data of the pipeline under test.

[0157] The data processing module is used to input the full matrix data of the pipeline under test into the trained autoencoder for processing, and obtain the defect echo signal of the pipeline under test.

[0158] Example 4

[0159] This embodiment provides a computer-readable storage medium storing a computer program that, when executed, implements the pipeline ultrasonic multimodal separation method described in Embodiment 1.

[0160] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0161] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0162] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0163] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for ultrasonic multimodal separation in pipelines, characterized in that, include: The pipeline under test is scanned using a phased array ultrasonic testing system to obtain full matrix data of the pipeline under test; The full matrix data of the pipeline under test is input into a trained autoencoder for processing to obtain the defect echo signal of the pipeline under test.

2. The pipeline ultrasonic multimodal separation method according to claim 1, characterized in that, The steps for obtaining the trained autoencoder include: Obtain the pre-set experimental parameters, including: phased array detection parameters, geometric condition parameters of defects and boundaries, and specimen parameters; the phased array detection parameters include: center frequency, number of array elements N, wedge angle α, and wedge height Z. s The distance between the front edge of the wedge and the edge; the geometric parameters of the defect and the boundary include: defect size, defect location, wedge interface, and bottom boundary of the specimen; the specimen parameters include: specimen thickness, specimen diameter, specimen density, transverse / longitudinal wave velocity of the specimen, weld shape, and weld size; By simulating real detection conditions through random perturbation of grid points, some experimental parameters are gridded, and random perturbation is added to each grid point to obtain experimental parameters with perturbation. The perturbation-containing experimental parameters are input into the constructed ultrasonic wave response models with different modes and the boundary response models with different modes to obtain full matrix data; the full matrix data includes ultrasonic wave information of different modes at the boundary and the defect, as well as the corresponding multiple echo information. Preprocess the full matrix data to obtain the preprocessed full matrix data; Construct a training set containing experimental parameters and preprocessed full matrix data, and use the defect locations in the experimental parameters as the predicted labels for the training set; The training set is input into the autoencoder for training to obtain the defect locations detected in the specimen; The autoencoder is trained by using the predicted labels from the training set and the defect locations detected in the specimen, and by minimizing the objective loss function to determine the optimal model parameters.

3. The pipeline ultrasonic multimodal separation method according to claim 2, characterized in that, The method of simulating real detection conditions by randomly perturbing grid points involves gridding some experimental parameters and adding random perturbations to each grid point to obtain perturbed experimental parameters, including: S01: Determine the range of a parameter based on the value of a certain parameter among some experimental parameters; the certain parameters are specimen thickness, wedge angle, and defect size. S02: Create a step size based on this parameter range. The structured mesh is used to obtain multiple mesh points; S03: Based on step size And grid points, calculate the random perturbation for each grid point; S04: Add a random perturbation at each grid point to obtain grid points with random perturbation; Based on steps S01-S04, partial experimental parameters with random perturbations are obtained, and based on these partial experimental parameters with random perturbations, perturbation-containing experimental parameters are constructed.

4. The pipeline ultrasonic multimodal separation method according to claim 3, characterized in that, The formula for calculating the random perturbation is as follows: ； in, Let m be the m-th grid point; For the random perturbation of the m-th grid point, This indicates the creation of two completely independent subsets from a network with random perturbations; The step size for the structured grid points.

5. The pipeline ultrasonic multimodal separation method according to claim 2, characterized in that, The preprocessing of the full matrix data to obtain preprocessed full matrix data includes: The full matrix data is dimensionality reduced, and the dimensionality-reduced full matrix data is obtained. The dimensionality-reduced full matrix data is linearly normalized so that the value range is kept within [0, 1], resulting in the preprocessed full matrix data.

6. The pipeline ultrasonic multimodal separation method according to claim 2, characterized in that, The different ultrasonic wave modes and defect response models are represented as follows: ； ； ； in, The excitation signal emitted by the i-th element; Indicates angular frequency; Indicates the location of the defect; The ultrasonic wave propagation modes are represented by LL, LT, TL, LT, TT, LL-L, LL-T, TT-L, TT-T, LT-L, LT-T, TL-L, and TL-T, where T represents transverse wave transmission, L represents longitudinal wave transmission, -T represents transverse wave reflection, and -L represents longitudinal wave reflection. The path time from the excitation of the i-th element to the reception of the j-th element; l is the imaginary unit; The path from the excitation of the i-th element to the reception of the j-th element. Scattering coefficient at; , These are the transmission coefficients of the excitation path of the i-th element and the receiving path of the j-th element through different media layers, respectively. , These are the reflection coefficients of the excitation path of the i-th element and the receiving path of the j-th element through different media layers, respectively. , Let be the diffusion attenuation coefficients of ultrasonic wave propagation on the excitation path of the i-th element and the receiving path of the j-th element, respectively. Let be the material attenuation coefficient for ultrasonic wave propagation along the excitation path of the i-th element and the receiving path of the j-th element. , These represent the far-field directivity of the i-th element excitation path and the j-th element receiving path under different ultrasonic wave propagation modes. For the i-th element and The included angle at the point; and These are the propagation processes of the excitation of the i-th element and the reception of the j-th element, respectively. in, ; ； in, The interface position along the k-path of ultrasonic wave propagation; Let be the angle of incidence at the interface along the k-path of ultrasonic wave propagation; The angle of refraction or reflection at the interface along the k-path of ultrasonic wave propagation; Let k be the speed of ultrasonic wave propagation along path k. Let n be the refractive index at the interface along the ultrasonic wave propagation path k; k is the ultrasonic wave propagation path, and n is the number of transmissions. Different incident waves, including incident transverse waves and incident longitudinal waves; in, ; ； ； ； ； ； ； Wherein, LL and LT represent longitudinal wave transmission and transverse wave transmission under longitudinal wave incident conditions, respectively; TL and TT represent longitudinal wave transmission and transverse wave transmission under transverse wave incident conditions, respectively. The transmission coefficients of longitudinal and transverse waves under longitudinal wave incidence conditions; The transmission coefficients of longitudinal and transverse waves under transverse wave incidence conditions; These are the angles between the incident longitudinal wave and the normal, and the angles between the incident transverse wave and the normal, respectively. These are the refraction angles of the longitudinal and transverse waves, respectively, under the condition of longitudinal wave incidence. These are the refraction angles of the longitudinal and transverse waves under the condition of transverse wave incidence, respectively; These are the reflection angles of the transverse wave under longitudinal wave incidence and transverse wave incidence conditions, respectively. These are the material densities of the specimen and the wedge block, respectively. These are the longitudinal wave velocity of the specimen, the transverse wave velocity of the specimen, the longitudinal wave velocity of the wedge, and the transverse wave velocity of the wedge, respectively. in, ; ； ； ； ； ； in, These represent longitudinal wave reflection and transverse wave reflection under longitudinal wave incidence conditions, respectively. These represent longitudinal wave reflection and transverse wave reflection under transverse wave incidence conditions, respectively. The reflection coefficients of longitudinal and transverse waves under longitudinal wave incidence conditions; The reflection coefficients of longitudinal and transverse waves under transverse wave incidence conditions; These are the angles between the sensor normal and the incident longitudinal wave, and the angles between the sensor normal and the incident transverse wave, respectively. These are the reflection angles of the transverse wave under longitudinal wave incidence and transverse wave incidence conditions, respectively. These are the longitudinal wave velocity and the transverse wave velocity, respectively.

7. The pipeline ultrasonic multimodal separation method according to claim 6, characterized in that, The different ultrasonic wave modes and boundary response models are used to represent the defect locations in the different ultrasonic wave modes and defect response models. Replace with a vector representing the position of the ultrasonic reflection point on the boundary. constitute.

8. The pipeline ultrasonic multimodal separation method according to claim 2, characterized in that, The autoencoder includes an encoder and a decoder connected in sequence; the encoder includes eight three-dimensional convolutional blocks connected in sequence; the decoder includes one three-dimensional convolutional block.

9. A pipeline ultrasonic multimodal separation system, characterized in that, include: The data acquisition module is used to scan the pipeline under test using a phased array ultrasonic testing system to obtain the full matrix data of the pipeline under test. The data processing module is used to input the full matrix data of the pipeline under test into the trained autoencoder for processing, and obtain the defect echo signal of the pipeline under test.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed, implements the pipeline ultrasonic multimodal separation method according to any one of claims 1-8.

Images