Uniform thinning detection method for pressure pipeline based on multi-modal guided wave fusion

Abstract

The invention discloses a pressure pipeline uniform thinning detection method based on multi-mode guided wave fusion. A composite guided wave excitation device is designed to synchronously excite multiple guided wave modes, and high-precision and long-distance wall thickness reduction detection is realized by combining a time-frequency domain conjoint analysis technology and a deep residual network of deep learning.

Description

Technical Field

[0001] This invention relates to the field of nondestructive testing technology, specifically to a method for detecting uniform thinning of pressure pipeline wall thickness based on multimodal guided wave mode fusion, which is particularly suitable for the inspection of in-service pressure pipelines in the fields of petrochemicals, energy and power. Background Technology

[0002] Pressure pipelines are crucial infrastructure for industrial production and energy transmission, and their safe operation is paramount. However, due to corrosion, erosion, and other factors, pipeline walls gradually thin, which can lead to leaks or even explosions in severe cases. Traditional inspection methods, such as ultrasonic testing and radiographic testing, can detect localized defects, but they lack the sensitivity to detect uniformly thinned pipelines and are difficult to implement over long distances.

[0003] Guided wave testing technology has gradually become an important means of pressure pipeline inspection due to its advantages such as long propagation distance and high detection efficiency. However, existing guided wave testing technologies are mainly aimed at local defects (such as cracks and holes), and lack sufficient sensitivity for detecting uniformly thinned defects. In addition, single-mode guided waves have a limited response characteristic to wall thickness changes and are easily affected by environmental noise. Traditional time-domain signal analysis methods are difficult to effectively extract minute thickness change features, and existing equipment cannot achieve a combination of long-distance detection and high-precision measurement. Summary of the Invention

[0004] 1. Technical problem: The present invention aims to overcome the problems of low sensitivity, poor signal-to-noise ratio and insufficient quantitative accuracy of existing guided wave detection technology for uniform thinning detection, and to provide a new detection method that integrates multi-mode guided wave characteristics.

[0005] 2. Technical Solution

[0006] 2.1 Composite Waveguide Excitation Module A wideband electromagnetic acoustic transducer (EMAT) array is used; Simultaneous excitation of L(0,2) and T(0,1) mode guided waves; Designed frequency range: 50 kHz - 150 kHz.

[0007] 2.2 Signal Processing System Core innovation: Method for constructing the joint time-frequency distribution matrix A high-dimensional feature matrix is ​​established through joint time-frequency domain analysis of multi-mode guided wave signals. The specific implementation steps are as follows: Step 1: Single-mode time-frequency analysis (1) Short-time Fourier Transform (STFT) Time-frequency decomposition of the L(0,2) mode signal:

[0008] Parameter settings: Window function: Hamming window, window length 256 sampling points Overlap rate: 50% Frequency resolution: Time and frequency accuracy of ±0.5kHz for a 100kHz signal. (2) Wavelet Packet Decomposition (WPD) Multi-resolution decomposition of the T(0,1) mode signal:

[0009] Parameter settings: Wavelet base: db4.

[0010] Number of decomposition layers: 6.

[0011] Effective frequency band: 3-6 layers of nodes (covering 50-150kHz).

[0012] Step 2: Time-Frequency Feature Fusion (1) Normalization

[0013] (2) Multimodal weighted fusion

[0014] The weighting coefficients were determined through experimental optimization. The objective function for optimization is:

[0015] (3) Nonlinear enhancement

[0016] Step 3: Matrix Dimensionality Reduction Principal component analysis (PCA) was used to retain 95% of the energy characteristics.

[0017] 2.3 Deep Learning Intelligent Analysis Input the feature matrix into the pre-trained ResNet-18 model and output the predicted wall thickness reduction value; Network architecture: An improved version of ResNet-18; Convolutional modules: 5, kernel size 3×3; Cross-layer connections: Add cross-layer connections after each convolutional module; Adaptive pooling: Global average pooling.

Claims

1. A method for detecting uniform thinning of pressure pipelines based on multimodal guided wave fusion, characterized in that... include: The process involves steps such as multimodal guided wave cooperative excitation, joint time-frequency domain feature extraction, and intelligent analysis based on deep learning.

2. The method according to claim 1, characterized in that... The waveguide mode combination includes, but is not limited to, L(0,2) and T(0,1) modes, with a frequency range of 50-150kHz.

3. The method according to claim 1, characterized in that... The construction of the joint time-frequency distribution matrix includes: Dual-mode processing of short-time Fourier transform and wavelet packet decomposition; Weighted fusion based on joint optimization of signal-to-noise ratio and sensitivity; Logarithmic augmentation and PCA dimensionality reduction combined operation.

Images