A high-precision underwater pipeline defect detection method

Abstract

This invention relates to a high-precision method for detecting defects in underwater pipelines, particularly a method combining eddy current detection (ECT) and alternating electromagnetic field detection (ACFM) technologies. The method involves installing a composite sensor on the surface of the underwater pipeline to collect eddy current and magnetic field signals. Signal processing methods such as filtering, Fourier transform, and wavelet denoising are used to process and extract features from the collected raw signals, yielding characteristic information reflecting pipeline defects. Then, a weighted data fusion algorithm is employed to fuse the detection results of ECT and ACFM, improving the accuracy of defect identification. Finally, a trained artificial intelligence model is used to classify and quantitatively analyze the fused features, outputting the type, location, and severity of the defects. This invention enables high-precision detection of cracks, corrosion, and weld defects, is suitable for pipeline inspection in complex underwater environments, and exhibits high detection accuracy and reliability.

Description

Technical Field

[0001] This invention relates to the field of underwater nondestructive testing technology, and more particularly to a high-precision method for detecting defects in underwater pipelines that combines eddy current testing (ECT) and alternating electromagnetic field testing (ACFM). This method is applicable to the detection and assessment of corrosion defects and cracks in underwater metal pipelines such as subsea oil and gas pipelines. Background Technology

[0002] Subsea pipelines are critical infrastructure for marine oil and gas transportation. Operating for extended periods in complex marine environments, their walls are prone to defects due to corrosion and fatigue. Conventional underwater pipeline inspection methods include ultrasonic testing, magnetic particle testing, and magnetic flux leakage testing. However, these methods are limited in their underwater applications by coupling media, surface deposits, and testing efficiency. Eddy current testing (ECT), an electromagnetic induction non-destructive testing technique, offers advantages such as non-contact operation, the ability to detect through coatings, and sensitivity to surface and near-surface defects. However, single-frequency ECT has limited accuracy in detecting defects at deeper locations or with complex shapes. Alternating current electromagnetic field testing (ACFM) is a new electromagnetic testing technology applied to underwater structures in recent years. It detects surface-opening defects such as cracks by introducing an alternating current onto the metal surface and measuring the magnetic field response. ACFM offers advantages such as accurate quantitative assessment of cracks and the ability to detect defects without removing coatings. However, its ability to detect non-opening volumetric defects (such as internal corrosion pits) is relatively insufficient. A single testing technology cannot comprehensively address the detection accuracy of all types of defects. To improve the reliability and accuracy of underwater pipeline defect detection, there is an urgent need for a data processing method that integrates multiple detection techniques to achieve high-precision defect identification by leveraging their complementary advantages. Summary of the Invention

[0003] In view of this, the present invention provides a high-precision detection method for underwater pipeline defects to solve the problems of insufficient accuracy and adaptability of single detection methods in the prior art. This method integrates two electromagnetic detection methods, ECT and ACFM, and achieves automatic identification of pipeline defects by preprocessing, extracting features, and fusing information from the two sensor signals, combined with artificial intelligence algorithms. First, the electromagnetic response signals of the pipeline surface are acquired using ECT and ACFM sensors; then, noise reduction filtering and time-frequency domain analysis are performed to extract parameters reflecting defect characteristics; next, the feature information of ECT and ACFM is fused using a fusion algorithm to form a more reliable defect indication; finally, a pre-trained intelligent discrimination model is used to classify the fused features and output the defect detection results. Compared with single methods, the method of the present invention can significantly improve the accuracy and detection rate of defect detection, has good detection effects on different types of defects (such as cracks and corrosion), and has the advantages of strong anti-noise interference capability and real-time automatic discrimination. Attached Figure Description

[0004] Figure 1 This is a flowchart illustrating the high-precision underwater pipeline defect detection method of the present invention. Detailed Implementation Plan

[0005] The following is in conjunction with the appendix Figure 1 (Flowchart) This invention provides a detailed description of a high-precision underwater pipeline defect detection method. The method integrates eddy current detection (ECT) and alternating electromagnetic field detection (ACFM), and involves detailed signal processing methods and mathematical formulas to accurately identify underwater pipeline crack defects, corrosion defects, and weld defects.

[0006] The specific implementation method includes the following steps:

[0007] Step 1, Data Acquisition: The composite detection device integrating the eddy current detection probe and the AC electromagnetic field detection probe is placed close to or suspended on the surface of the underwater pipe to be tested, and the eddy current signal and magnetic field distortion signal are collected respectively.

[0008] Step two, signal preprocessing step, involves using the following signal processing methods for denoising and filtering of the eddy current signal and ACFM signal:

[0009] Specifically, the acquired raw signal X(t) is bandpass filtered, and the expression for the bandpass filter is as follows:

[0010]

[0011] Where X(t) is the original time-domain signal, h(t) is the impulse response of the bandpass filter, and the filtering frequency range is determined according to the detection depth.

[0012] Furthermore, wavelet thresholding is used for noise reduction, and the processing formula is as follows:

[0013]

[0014] Where W(·) represents wavelet transform, W -1 T represents the inverse wavelet transform. λ This represents the threshold processing function, where λ is the threshold value. Specific optional soft threshold processing functions include:

[0015] T λ (x) = sign(x)max(|x|-λ,0)

[0016] Step 3, signal frequency domain transformation and feature extraction: The preprocessed ECT and ACFM signals are processed using Fast Fourier Transform (FFT), with the specific formula as follows:

[0017]

[0018] The amplitude feature A(f) and phase feature φ(f) are extracted from the spectrum, specifically as follows:

[0019]

[0020] Where Re[X(f)] and Im[X(f)] are the real and imaginary parts of the spectrum, respectively.

[0021] Simultaneously, the time-domain features of ACFM are extracted, including signal B. x Valley depth numerical R sx and signal B z The height difference between the peaks and troughs R sz The specific calculations are as follows:

[0022] R sx =|Bx max -Bx min |

[0023] R sz =|Bz max -Bz min |

[0024] Step four, signal fusion processing step, employing a feature-level data fusion method to fuse the features of the ECT and ACFM signals to construct a joint feature vector F, using a weighted fusion algorithm:

[0025] F = w1·F ECT +w2·F ACFM

[0026] Among them, F ACFM =[R sx ,R sz ], F ECT =[A(f),φ(f)];

[0027] w1 and w2 are the fusion weights of ECT and ACFM, respectively, which are dynamically adjusted according to the actual situation of the detection area, and satisfy w1+w2=1.

[0028] Furthermore, based on the different sensitivities of cracks and corrosion, weight values ​​are determined. Specifically, ACFM has a larger weight when detecting crack defects; ECT has a larger weight when detecting corrosion defects; and the weights are equal or adjusted according to the actual situation when detecting weld areas.

[0029] Step 5, Defect Identification and Classification: The fused features are input into a pre-trained artificial intelligence discrimination model. Specifically, the model uses a deep convolutional neural network (CNN), whose structure includes convolutional layers, pooling layers, and fully connected layers. The output layer uses the Softmax function to determine the defect category.

[0030] The output of a neural network is defined as:

[0031]

[0032] Where x represents the input fusion feature; z j θ is the linear combination input of the j-th node of the output layer; θ is the network weight parameter; K is the number of defect types, and the output classifies and outputs the type and severity of defects such as cracks, corrosion or welds.

[0033] Step six, quantitative defect assessment: Based on the identified defect types, further quantitative analysis is performed. For crack defects, the crack length L and depth D are estimated using the established two-parameter joint probability density function based on the ACFM butterfly signal characteristics Bx and Bz.

[0034] The crack size probability density function is defined as follows:

[0035]

[0036] Where: μ sx ,μ sz σ is the average value of the standard crack characteristic signal. sx ,σ sz ρ represents the corresponding standard deviation, and ρ is the correlation coefficient.

[0037] For corrosion defects, a corrosion model is established based on the amplitude variation of the ECT signal spectrum, and the depth of the corrosion defect is calculated through the amplitude variation of characteristic spectral lines: D corr =k·ΔA(f)+b

[0038] In the formula, k and b are linear coefficients calibrated through experiments; ΔA(f) is the change in amplitude of the ECT signal.

[0039] Step six, the results feedback step, transmits the above identification results and quantitative assessment results to the control terminal, displays the type and specific location of pipeline defects, and provides subsequent maintenance personnel with accurate repair work.

[0040] The above method enables high-precision underwater pipeline defect detection and improves the reliability and accuracy of identifying and quantitatively analyzing different types of defects.

[0041] The above embodiments are for detailed description of the present invention and not for limitation. Any changes and modifications made by those skilled in the art based on the teachings of this invention are within the scope of protection of the claims of this invention.

Claims

1. A high-precision detection method for underwater pipeline defects, characterized in that, Includes the following steps: Step 1, Data Acquisition. This is achieved using an eddy current testing (ECT) sensor and an AC electromagnetic field detector. The Alternating Current Field Measurement (ACFM) sensor scans the surface of an underwater pipeline to acquire raw detection signals of pipeline defects in a non-contact state; the ECT sensor generates an alternating electromagnetic field to induce eddy currents in the pipe wall, and the acquired eddy current signals reflect the changes in the electromagnetic response of the pipeline's metallic material; the ACFM sensor introduces an alternating current field into the pipeline surface to acquire alternating magnetic field signals on the pipeline surface, where the presence of defects will cause magnetic field distortion signals. Step 2, Signal Preprocessing. The raw detection signals acquired by ECT and ACFM are filtered to eliminate underwater environmental noise and irrelevant interference. Digital filtering algorithms are used to smooth and denoise the signals. The filtering process is implemented through convolution operations, for example: Where x[n] is the original discrete detection signal, y[n] is the filtered signal, h[k] is the impulse response coefficient of the filter, and M is the filter order. By selecting appropriate filter coefficients h[k], high-frequency noise and background interference signals are filtered out, thereby improving the signal-to-noise ratio. Step 3, Signal Transformation and Feature Extraction. Fourier transforms are performed on the filtered ECT and ACFM signals respectively, converting them from the time domain to the frequency domain to extract spectral features. The Fourier transform satisfies: Where x(t) is the time-domain function of the input signal, and X(f) is its frequency-domain spectrum. The amplitude and phase distributions of each signal in the frequency domain are obtained through the above transformation, and characteristic parameters including the dominant frequency amplitude, harmonic component amplitude ratio, phase difference, signal peak value, and zero drift are extracted. For the ACFM signal, the characteristics of the distorted magnetic field are further extracted, such as the peak-to-valley difference of the magnetic field component along the crack direction and the step change value of the magnetic field component perpendicular to the crack direction; for the ECT signal, the changes in the real and imaginary parts of the coil's equivalent impedance are extracted. Step 4, Data Fusion. Based on the extracted ECT and ACFM feature parameter sets, a data fusion algorithm is used to fuse features and construct a joint feature vector for defect identification. The data fusion employs an algorithm combining weighted fusion and decision reasoning: features extracted by each detection method are assigned predetermined weights, the ECT feature vector vECT and the ACFM feature vector vACFM are merged, and the fused feature vector is calculated. vF=[w1vECT,;w2v ACFM ] Where w1 and w2 are weighting coefficients (satisfying w1+w2=1). Optionally, the defect indications of ECT and ACFM can be fused at the decision level using evidence reasoning methods, and the confidence levels corresponding to the two detection results can be synthesized using Dempster-Shafer theory to improve the reliability of defect judgment. Step 5, Defect Identification and Assessment. The fused features are input into a pre-trained AI defect discrimination model to automatically identify and quantitatively assess pipeline defects. The AI ​​discrimination model is implemented using machine learning algorithms, such as classifiers based on neural networks or support vector machines. Model parameters are obtained by training on a large amount of known defect data, and a nonlinear mapping function f(x) is constructed to map feature vectors to defect states. The model outputs defect identification results for the input fused feature vector f(x). The discrimination model outputs the defect category (e.g., cracks, corrosion, or other defects) and a defect severity assessment index. When the output of f(x) exceeds a set threshold, a defect is determined to exist. The defect size parameters are estimated by combining the quantitative relationship between ECT and ACFM features, achieving high-precision detection and assessment of underwater pipeline defects.

2. The method according to claim 1, characterized in that, The ECT data acquisition step employs multi-frequency eddy current sensing technology, simultaneously applying at least two different frequencies of excitation current to excite eddy currents, thereby obtaining response signals at different penetration depths and improving the detection sensitivity of defects at different depths in pipelines.

3. The method according to claim 1, characterized in that, The ACFM data acquisition step uses a dual orthogonal coil probe to simultaneously measure the alternating magnetic field components along the axial and radial directions of the pipe, where the magnetic field signal B along the defect direction is... x Magnetic field signal B perpendicular to the defect orientation z Used to construct a butterfly-shaped feature map to characterize the length and depth of surface cracks.

4. The method according to claim 1, characterized in that, The signal preprocessing step uses a wavelet denoising algorithm to preprocess the ECT and ACFM signals, removing sudden interference and background noise while retaining the effective defect signal components. The wavelet denoising includes multi-scale wavelet decomposition, threshold filtering, and reconstruction of the signal.

5. The method according to claim 1, characterized in that, The feature extraction step further includes performing statistical feature extraction on the time-domain signal to obtain time-domain features such as the mean amplitude, root mean square value, kurtosis, and margin of the signal, and combining them with the frequency-domain features to form a comprehensive feature vector to enrich the feature information for defect identification.

6. The method according to claim 1, characterized in that, The data fusion algorithm adopts a fuzzy inference fusion strategy, which standardizes the ECT and ACFM features based on the membership function, and obtains the fusion output through fuzzy rule inference, thereby improving the robustness of defect identification when there is information uncertainty.

7. The method according to claim 1, characterized in that, The artificial intelligence defect discrimination model is a deep neural network, which includes several hidden layers to learn the nonlinear relationship between ECT and ACFM features and defect types. The deep neural network model is trained by the backpropagation algorithm and outputs the probability value of the existence of defects during the recognition process, thereby realizing the automatic judgment of defects.

8. The method according to claim 1, characterized in that, During the detection process, the data and results obtained in the above steps are stored in the database, and the artificial intelligence discrimination model is updated online using a continuous learning algorithm, so that the model can continuously improve the recognition accuracy as new defect samples are added.

Images