Ultrasonic identification method for cold welding defects of gas PE pipeline electric fusion joint

Abstract

This invention discloses an ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines, specifically relating to the field of non-destructive testing technology. The method involves circumferential ultrasonic scanning of the electrofusion joint, acquiring one-dimensional radio frequency time-domain ultrasonic signals and performing noise reduction. The denoised signals are then reconstructed into two-dimensional ultrasonic grayscale images, and three-zone division and differentiated lightweight processing are implemented. Four types of features are extracted, quantized, encoded, and stored. A multi-feature hierarchical fusion mechanism is used to accurately identify cold welding defects, distinguishing between defects such as pores, oxide impurities, and incomplete socket insertion, and implementing gradient risk classification. Based on the risk level, a rapid inspection and precise re-inspection mode is adaptively switched to complete secondary discrimination, ultimately generating a traceable identification report. This method addresses the problem of high false positive rates caused by noise interference from material scattering and resistance wire clutter in traditional methods, improving the signal-to-noise ratio and identification accuracy of weak cold welding signals.

Description

Technical Field

[0001] This disclosure relates to the field of pipeline non-destructive testing technology, and more specifically to an ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines. Background Technology

[0002] Polyethylene (PE) gas pipelines are widely used in gas transmission and distribution systems due to their advantages such as corrosion resistance, high flexibility, and convenient construction. Electrofusion welding is the mainstream process for on-site pipeline connection, and the fusion quality of the joints directly affects the operational safety of the pipeline network. Cold welding is a frequent and common defect in electrofusion construction. It originates from insufficient melting or rapid cooling of the weld surface, resulting in a continuous, slightly unfused thin layer at the fusion interface. This defect only has a weak difference in acoustic impedance and no obvious structural echo characteristics. Under long-term pipeline load and medium pressure, cold welding defects can easily expand gradually, leading to joint cracking and gas leakage, which is a major safety hazard to the pipeline network. Therefore, achieving rapid and accurate detection of cold welding defects is a key technical requirement that urgently needs to be addressed in the field of gas pipeline operation and maintenance.

[0003] Currently, the commonly used detection methods in existing technologies mainly rely on traditional ultrasonic testing and manual interpretation, which have significant shortcomings. Traditional single-crystal A-scan ultrasonic testing has extremely low sensitivity to weak interface defects in cold welds. It is severely affected by sound attenuation from PE materials, particle scattering noise, and strong reflection clutter from resistance wires. Weak incomplete fusion signals are easily drowned out, and effective identification based solely on one-dimensional waveforms is difficult, resulting in a high rate of missed detections. Furthermore, it relies entirely on human experience for interpretation, leading to significant misjudgments. The classic phased array spacing quantification method achieves quantitative evaluation of cold welds by extracting the distance between feature lines and resistance wires, and is the mainstream benchmark method in the industry. However, this method is highly dependent on manual image interpretation and distance measurement, resulting in low detection efficiency, large human error, and the use of only a single geometric parameter to determine defects. It cannot effectively distinguish between similar near-interface defects such as cold welds, fusion voids, oxide inclusions, and incomplete socket insertion, leading to significant misjudgments. Summary of the Invention

[0004] This disclosure provides an ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines, which achieves high-precision, low-false-judgment, and hierarchical intelligent identification of cold welding defects under complex working conditions, while taking into account the needs of large-scale rapid inspection and accurate re-inspection of high-risk defects, thereby improving the safety and detection efficiency of gas pipeline operation and maintenance.

[0005] An ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines, comprising: S1. Perform a circumferential scan on the electrofusion joint of the gas PE pipeline to obtain the one-dimensional radio frequency time domain ultrasonic signal of the electrofusion joint, and perform adaptive noise reduction processing on the one-dimensional radio frequency time domain ultrasonic signal to suppress the interference of polyethylene material scattering noise, resistance wire metal reflection clutter and probe coupling fluctuation noise. S2. Generate a two-dimensional ultrasonic grayscale image based on the noise-reduced one-dimensional radio frequency time-domain ultrasonic signal, and divide the two-dimensional ultrasonic grayscale image into the outer wall background region, the resistance wire feature region and the fusion interface region, and perform regional differentiated lightweight processing. S3. Extract geometric spacing features, interface echo amplitude features, fusion zone texture uniformity features, and defect circumferential continuous distribution features from the lightweight processed two-dimensional ultrasonic grayscale image, and perform differential fixed-point quantization encoding and storage on the extracted features. S4. A multi-feature hierarchical fusion discrimination mechanism is adopted to identify cold welding defects and distinguish them from point-like fusion holes, pipe wall oxidation impurities and pipe socket defects. Based on the circumferential distribution ratio of abnormal areas of the fusion interface and the degree of feature offset, the gradient risk classification of cold welding defects is carried out, and the multi-level evaluation results of cold welding risk are output. S5. Based on the multi-level assessment results of cold welding risks, adaptively switch between rapid inspection and precise re-inspection modes to perform secondary judgment and generate a cold welding defect identification report.

[0006] Furthermore, the step of performing a circumferential scan of the electrofusion joint of the gas PE pipeline to obtain the one-dimensional radio frequency time-domain ultrasonic signal of the electrofusion joint, and performing adaptive noise reduction processing on the one-dimensional radio frequency time-domain ultrasonic signal, includes: The 0° longitudinal wave linear phased array probe is driven to fit against the outer wall surface of the electrofusion joint and performs a 360° full-coverage circumferential scan along the circumference of the electrofusion joint at a constant angular velocity. During the scan, ultrasonic beams are continuously and directionally emitted into the electrofusion joint and ultrasonic simulated echo signals reflected by the medium inside the electrofusion joint are received. The received ultrasonic analog echo signal is converted from analog to digital by the multiplexing channel in the phased array instrument to generate a one-dimensional radio frequency time domain ultrasonic signal sequence. The one-dimensional radio frequency time domain ultrasonic signal sequence contains the original time domain information of polyethylene material scattering, reflection of the built-in resistance wire of the electrofusion joint and reflection of the fusion interface. Based on the diameter, pitch, and initial phase information of the built-in resistance wire in the electrofusion joint, a resistance wire position mask is formed; Outside the area covered by the resistive wire position mask, a sliding median filter is applied to the one-dimensional radio frequency time-domain ultrasonic signal sequence along the circumferential scanning path to obtain a dynamic noise substrate, which is used to reflect the inherent scattering characteristics of polyethylene (PE) material. The dynamic noise floor is subtracted point by point from the one-dimensional radio frequency time-domain ultrasound signal sequence to obtain the differential signal, and the wavelet soft threshold function is used to filter the obtained differential signal to remove global scattering noise. Output a noise-reduced one-dimensional radio frequency time-domain ultrasonic signal.

[0007] Furthermore, the generation of a two-dimensional ultrasound grayscale image based on the denoised one-dimensional radio frequency time-domain ultrasound signal includes: The beamforming method is used to perform coherent superposition reconstruction of the noise-reduced one-dimensional radio frequency time-domain ultrasonic signal. The beamforming method performs phase compensation and energy accumulation on radio frequency time-domain ultrasonic signals at different propagation depths, and coherently superimposes the acoustic beam energy at the same depth position to generate a two-dimensional ultrasonic grayscale image. The two-dimensional ultrasonic grayscale image is used to characterize the internal cross-sectional structure of the electrofusion joint. In a two-dimensional ultrasonic grayscale image, the horizontal axis coordinate maps to the circumferential scanning position, and the vertical axis coordinate maps to the propagation depth of the ultrasonic beam in the direction of the wall thickness of the electrofusion joint. The grayscale value of the pixel in the two-dimensional ultrasonic grayscale image is positively correlated with the acoustic impedance difference and ultrasonic reflection coefficient of the medium at the corresponding position.

[0008] Furthermore, the step of dividing the two-dimensional ultrasonic grayscale image into an outer wall background region, a resistance wire feature region, and a fusion interface region, and performing differentiated lightweight processing for each region, includes: The mean vertical grayscale value is calculated column by column for the denoised two-dimensional ultrasound grayscale image to obtain the vertical grayscale projection curve, and the vertical segmentation boundary of the region is determined based on the numerical abrupt change point of the vertical grayscale projection curve. Pixel mask annotation is performed on the denoised two-dimensional ultrasonic grayscale image based on the longitudinal segmentation boundary of the region, generating three sets of pixel masks. The three sets of pixel masks correspond to the outer wall background region, the resistance wire feature region, and the fusion interface region, respectively. Based on the pixel masks, the data of different regions are diverted to different computing links. For the outer wall background area, perform pixel downsampling and background color recognition, and disable edge detection, grayscale projection and texture parsing operations, and output the grayscale reference parameters of the area; For the resistance wire feature region, perform contour extraction, output the resistance wire boundary coordinate array, and disable global texture statistics operation; For the fusion interface region, downsampling and pixel simplification operations are disabled, and bilateral filtering, grayscale projection and edge extraction are performed. Full operations are performed on the pixel data in the region, and grayscale, texture and boundary parameters are output.

[0009] Furthermore, the pixel mask annotation is performed on the denoised two-dimensional ultrasonic grayscale image based on the longitudinal segmentation boundary of the region, generating three sets of pixel masks, including: A mapping storage table of vertical intervals and mask identifiers is pre-established. The mapping storage table records three sets of vertical coordinate intervals, and each set of vertical coordinate intervals is bound to an independent mask identifier. After reading the vertical coordinate values ​​corresponding to the pixels of the denoised two-dimensional ultrasound grayscale image, compare them with the vertical coordinate interval values ​​recorded in the mapping storage table. When the vertical coordinate value corresponding to a pixel is between the upper and lower limits of the vertical coordinate interval, the mask identifier bound to the vertical coordinate interval is retrieved and used as the mask identifier for matching the current pixel. When the vertical coordinate value corresponding to a pixel is outside the vertical coordinate range, it does not belong to the outer wall background area, the resistance wire feature area, or the core area of ​​the fusion interface. The preset blank mask identifier is called as the mask identifier for the current pixel matching. Write the retrieved mask identifier into the pixel attribute field corresponding to the current pixel to complete the single-pixel mask identifier matching operation.

[0010] Furthermore, the extraction of geometric spacing features, interface echo amplitude features, fusion zone texture uniformity features, and defect circumferential continuous distribution features, and the differential fixed-point quantization encoding and storage of the extracted features, includes: The difference conversion operation is performed on the skeleton coordinates of the characteristic region of the resistance wire and the effective pixel coordinates of the fusion interface, and the numerical conversion is performed in combination with the image resolution parameters to generate vertical spacing sub-features and circumferential spacing sub-features, which together constitute the geometric spacing features. Extract the grayscale statistical parameters of the effective pixels in the fusion interface region and combine them to generate interface echo amplitude features; The neighborhood gray-level co-occurrence matrix is ​​solved for the pixel matrix of the fusion interface region, and multiple sets of texture sub-features are extracted from the solved gray-level co-occurrence matrix to generate texture uniformity features of the fusion area. The circumferential extension length of the continuously distributed abnormal pixel clusters within the fusion interface region is statistically analyzed, and the ratio is converted based on the total length of the circumferential pixels of the fusion interface to generate the circumferential coverage ratio sub-feature, which constitutes the circumferential continuous distribution feature of the defect. For different types of features, an integer encoding format with an appropriate bit width is selected, and the actual physical value of each feature is normalized and mapped to the encoding value range corresponding to the integer encoding format to achieve fixed-point quantization encoding. The feature data generated by the encoding process is written into a pre-configured structured array, and storage address segments are allocated for various types of feature data in the structured array for storage.

[0011] Furthermore, the multi-feature hierarchical fusion discrimination mechanism performs a four-layer progressive screening of candidate regions to distinguish defect types, including: S401. Based on the interface echo amplitude characteristics and the circumferential continuous distribution characteristics of defects, perform initial screening of cold welding defects. When the interface echo amplitude characteristics are less than the first preset threshold and the circumferential continuous distribution characteristics of defects are greater than the second preset threshold, they are determined to be candidate areas for cold welding defects. S402. Based on the geometric spacing characteristics, the position compliance of the candidate cold welding defect areas after the initial screening is checked, and the deviation value between the geometric spacing characteristics and the preset standard spacing is calculated. When the deviation value exceeds the third preset threshold, it is removed, and the candidate cold welding defect areas whose positions conform to the physical laws of the fusion interface are retained. S403. Based on the texture uniformity characteristics of the fusion zone, perform texture compliance verification on the candidate regions of cold welding defects retained after position verification. When the texture uniformity characteristics are greater than the fourth preset threshold, remove them and retain the candidate regions of cold welding defects with non-uniform texture characteristics. S404. Based on the distribution characteristics of the resistance wire feature region, perform structural continuity verification on the candidate regions of cold welding defects retained after texture verification. When the candidate regions of cold welding defects exhibit local point-like anomalies and the resistance wire distribution remains continuous, they are removed, while candidate regions of cold welding defects accompanied by resistance wire distribution distortion or global texture deterioration are retained. S405. The candidate areas for cold welding defects retained after four-layer verification are determined as cold welding defect areas.

[0012] Furthermore, the method of classifying cold welding defects into gradient risks based on the circumferential distribution ratio and feature offset degree of abnormal regions at the fusion interface, and outputting multi-level evaluation results of cold welding risks, includes: The texture uniformity characteristics of the fusion zone and the circumferential continuous distribution characteristics of defects are weighted and calculated to obtain the cold welding confidence score. Based on the cold welding confidence score, the gradient risk classification of cold welding defects is carried out. When the confidence score of cold welding is less than the first classification threshold, it is judged as a low-risk cold welding defect; When the cold welding confidence score is greater than the first level threshold but less than the second level threshold, it is judged as a medium-risk cold welding defect. When the cold welding confidence score is greater than the second-level threshold, it is judged as a high-risk cold welding defect; Wherein, the second grading threshold is greater than the first grading threshold.

[0013] Furthermore, the adaptive switching between rapid inspection and precise re-inspection modes based on the multi-level assessment results of cold welding risks, and the subsequent secondary judgment, includes: Read the multi-level risk assessment results of cold welding and match the preset branch control logic to perform the switching operation between rapid inspection and precise re-inspection working modes; When the multi-level assessment result of cold welding risk is low-risk cold welding defect, switch to the rapid inspection mode, retrieve the geometric spacing features for preliminary screening, drive the scanning mechanism to complete the displacement according to the preset displacement step size, and switch to the next electrofusion joint to carry out cold welding defect identification. When the multi-level evaluation result of cold welding risk is medium-risk cold welding defect or high-risk cold welding defect, switch to the precise re-inspection mode, re-execute the circumferential scanning for the current spatial position of the electrofusion joint, obtain the ultrasonic signal, and call the full feature extraction algorithm for secondary discrimination. The full feature extraction algorithm has built-in bilateral filtering operation and gray-level co-occurrence matrix high-order statistical quantity solution operation. After completing the secondary discrimination, the output result of the secondary discrimination is compared with the multi-level evaluation result of cold welding risk. If the output result of the secondary discrimination is a low-risk cold welding defect, the cold welding risk evaluation result of the current electrofusion joint is updated to a low-risk cold welding defect, and the scanning mechanism is driven to complete the displacement according to the preset displacement step size, and switch to the next electrofusion joint to carry out cold welding defect identification.

[0014] Furthermore, the generation of the cold welding defect identification report includes: The one-dimensional radio frequency time-domain ultrasound signal, the two-dimensional ultrasound grayscale image, the extracted features, the multi-level evaluation results of cold welding risk and the secondary discrimination results are associated and bound together. Asymmetric encryption algorithm is used to encrypt the associated data to generate a cold welding defect identification report; The encrypted cold welding defect identification report is stored, and an index directory containing timestamps, geolocation information, and electrofusion joint numbers is created.

[0015] The beneficial effects of this invention are as follows: This invention employs a combined noise reduction scheme of dynamic noise substrate extraction under resistance wire position mask constraint and wavelet soft threshold filtering. It specifically achieves layered suppression of three types of interference: polyethylene material scattering noise, resistance wire metal reflection clutter, and probe coupling fluctuation noise. Without losing the effective echo characteristics of the fusion interface, it improves the signal-to-noise ratio of the ultrasonic signal, eliminates the hidden danger of misjudgment of cold welding defects caused by noise from the signal source, and solves the problem that the ultrasonic echo characteristics are easily submerged under the high scattering characteristics of PE pipes.

[0016] This invention uses pixel masking to precisely distribute two-dimensional ultrasonic grayscale images to different computation links. Only grayscale reference parameters are output for the outer wall background area, only boundary coordinates are extracted for the resistance wire feature area, and the full number of pixels are retained for computation in the fusion interface area. While preserving the core identification features of cold welding defects, the amount of computational data for the entire image is compressed, and the computation time for single-joint feature extraction is greatly shortened. This not only meets the needs of rapid inspection of a large number of joints on site, but also avoids the problem of loss of defect details caused by traditional global image compression.

[0017] This invention constructs a four-layer progressive multi-feature hierarchical fusion discrimination mechanism. Through the step-by-step filtering logic of amplitude screening, position verification, texture verification, and structural continuity verification, it achieves accurate differentiation between cold welding defects and three easily confused defects: point-like fusion holes, pipe wall oxidation impurities, and incomplete pipe socketing. This improves the accuracy of cold welding defect identification and solves the problem of discrimination confusion caused by the overlap of different defect features in on-site inspection.

[0018] This invention employs an adaptive switching mechanism between rapid inspection and precise re-inspection. In low-risk scenarios, initial screening can be completed solely based on geometric spacing features to advance to the next joint inspection. In medium- and high-risk scenarios, encrypted re-scanning and full-volume high-order feature calculation are automatically triggered. This balances the efficiency requirements of on-site batch inspection with the accuracy requirements for identifying high-risk defects. While ensuring a reduced rate of missed detection of cold welding defects, the overall inspection efficiency is improved compared to the traditional full-volume, full-joint high-precision inspection mode. It is particularly suitable for batch inspection scenarios of in-service electrofusion joints spanning hundreds of kilometers of urban gas pipeline networks. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A schematic diagram of the overall method flow of the present invention; Figure 2 This is a logical diagram of the multi-feature hierarchical fusion discrimination mechanism in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0021] To enable those skilled in the art to better understand the present application, embodiments of the present invention disclose an ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines, as shown in the appendix. Figure 1 As shown, it includes the following steps: S1, Circumferential scanning and adaptive noise reduction of ultrasonic signals The core function of this step is to complete the acquisition of the original ultrasonic signal with full coverage, and to specifically remove three types of interference: PE material scattering noise, resistance wire metal reflection noise, and probe coupling fluctuation noise. This maximizes the preservation of the effective echo signal of the fusion interface and solves the problems of traditional ultrasonic testing where the effective signal is submerged by noise, baseline drift, and numerous clutter peaks.

[0022] In practice, a 0° longitudinal wave linear phased array probe is selected to ensure that the ultrasonic waves are perpendicularly incident on the pipe wall, avoiding signal distortion and refraction attenuation caused by oblique incidence. The probe is tightly attached to the outer wall of the electrofusion joint, and a coupling agent is used to ensure uniform coupling throughout the entire area, eliminating fluctuation noise caused by poor local coupling. The scanning mechanism is controlled to perform a 360° continuous circumferential scan around the electrofusion joint at a constant angular velocity, achieving signal acquisition without blind spots in all fusion areas around the joint and avoiding local missed detections. During the scanning process, a high-frequency ultrasonic beam is continuously emitted directionally into the pipe fitting. The sound waves penetrate the PE outer wall, the resistance wire area, and the fusion interface, and are finally reflected at the interface of different media. The probe continuously receives the ultrasonic simulated echo signal returned by the internal media.

[0023] The phased array instrument uses its built-in multiplexing channel to perform high-speed analog-to-digital conversion on the continuous analog echo signal, discretizing and generating a complete one-dimensional radio frequency time-domain ultrasonic signal sequence. This signal sequence fully preserves all the original time-domain information of the microscopic scattered echo of the PE material, the strong reflected echo of the resistance wire metal, and the echo of the fusion interface between the pipe and fittings, providing original data support for subsequent image reconstruction and defect analysis.

[0024] To isolate interference from strong reflections of the resistance wire, this invention uses the factory parameters of the electrofusion joint under test, including the diameter, pitch, and initial phase parameters of the built-in resistance wire, and precisely generates a resistance wire position mask through geometric modeling. This mask can accurately pinpoint the interference area of ​​the resistance wire in the signal dimension, preventing the strong reflection signal from the resistance wire from being misjudged as a defect signal.

[0025] In the effective fusion detection area outside the resistance wire position mask, the one-dimensional radio frequency time-domain ultrasonic signal sequence is subjected to segment-by-segment sliding median filtering along the circumferential scanning time sequence path, dynamically fitting and generating a dynamic noise substrate adapted to the current pipe condition. Unlike traditional fixed noise thresholds, this dynamic noise substrate can adaptively change in real time with the inherent scattering characteristics of PE material, on-site temperature, and pipe aging state, accurately characterizing the background noise level under the current testing conditions.

[0026] The original one-dimensional radio frequency time-domain ultrasonic signal sequence is differentially analyzed point by point with a dynamic noise substrate to subtract the inherent scattering background of the material, resulting in a differential signal with baseline drift removed. Then, a wavelet soft thresholding function is used to perform global fine filtering on the differential signal. The wavelet soft thresholding algorithm can smooth random noise while preserving the signal abrupt change characteristics (i.e., the echo characteristics of fusion interface defects) to the greatest extent possible, effectively filtering out global random scattering noise and probe coupling fluctuation noise. The final output is a high signal-to-noise ratio, high-fidelity denoised one-dimensional radio frequency time-domain ultrasonic signal, achieving the technical effect of multi-source noise layered suppression and accurate preservation of effective features.

[0027] Specifically, the probe is set to a uniform scanning angular velocity of 5 mm / s. The detector's multiplexed channel performs 16-bit analog-to-digital conversion on the continuously acquired analog echoes to generate a complete one-dimensional radio frequency time-domain ultrasonic signal sequence. This sequence fully records three types of reflection signals: random scattered echoes from polymer particles in the PE pipe, strong reflected echoes from the spiral resistance wire metal, and reflected echoes from the fusion interface between the sleeve and the pipe.

[0028] The operator enters the current geometric parameters of the sleeve resistance wire into the equipment interface: diameter 0.8mm, pitch 3mm, initial phase 0°. The equipment generates a pixel-level resistance wire position mask based on the geometric model, accurately locking the time domain signal interval corresponding to all strong reflection signals of the resistance wire.

[0029] The filtering operation is performed only in the effective detection time domain outside the resistance wire position mask. The RF signal is subjected to a sliding median filter with a length of 21 points along the circumferential scanning time sequence. The real-time dynamic noise substrate is fitted time by time. This substrate only reflects the inherent scattering noise of the PE material itself and is not affected by the strong reflection peak of the resistance wire.

[0030] The original RF time-domain signal is subtracted point by point from the corresponding dynamic noise substrate to obtain the differential signal with the fixed scattering substrate removed. The db4 wavelet basis is selected to perform multi-scale decomposition on the differential signal, and an adaptive soft threshold function is used to filter out random high-frequency scattering noise. Finally, the noise-reduced RF time-domain signal with a signal-to-noise ratio improved by more than 60% is output, which completely solves the defect of traditional global filtering that will drown out the weak interface echo of cold welding.

[0031] S2, Lightweight Reconstruction and Differentiated Processing of Two-Dimensional Ultrasonic Grayscale Images The core function of this step is to transform a one-dimensional time-domain signal into a visualized two-dimensional structural image. Through intelligent partitioning and differentiated operation strategies, it achieves speed reduction and load reduction in the background area and high-precision analysis of the core fusion area, solving the problems of low efficiency and weakened core defect features in traditional unified operation.

[0032] In practice, an adaptive beamforming algorithm is used to coherently superimpose and reconstruct the one-dimensional radio frequency time-domain ultrasonic signal after S1 noise reduction. The algorithm performs precise phase compensation on the time-domain signals of ultrasonic waves at different propagation depths to eliminate image blurring caused by the propagation time difference of the sound beams; at the same time, it coherently superimposes the energy of multiple channels of sound beams at the same wall thickness depth, which greatly improves the imaging gain of the effective interface echo and reconstructs a high-resolution two-dimensional ultrasonic grayscale image.

[0033] The generated two-dimensional ultrasonic grayscale image has a strict physical mapping relationship: the horizontal axis of the image precisely corresponds to the angle position of the circumferential scanning of the electrofusion joint, realizing the circumferential defect location; the vertical axis corresponds to the propagation depth of the ultrasonic wave in the pipe wall thickness direction, realizing the defect depth location; the grayscale value of the image pixels is positively correlated with the acoustic impedance difference and ultrasonic reflection coefficient of the medium at the corresponding location. The acoustic impedance of the fusion-poor and cold-welded delamination areas changes abruptly, and the grayscale value is significantly abnormal. It can intuitively distinguish the intact fusion area from the defect area and accurately characterize the internal cross-sectional structure of the joint.

[0034] Specifically, the noise-reduced radio frequency time-domain signal output by S1 is read, and phase compensation delay is applied independently to the radio frequency signals at different wall thicknesses and propagation depths. The acoustic beam energies of all array elements at the same wall thickness depth are coherently superimposed to reconstruct a two-dimensional ultrasonic grayscale B-scan image. The horizontal axis of the image is proportionally mapped to the circumferential scanning displacement coordinates of the probe, and the vertical axis corresponds one-to-one with the wall thickness depth coordinates. The pixel grayscale value is positively correlated with the difference in acoustic impedance of the medium. The grayscale difference of the intact fusion interface is small, the grayscale fluctuation of the cold welded area is slight, and the metal resistance wire exhibits a high-brightness grayscale feature, indicating a clear physical imaging logic.

[0035] To achieve precise partitioning, the mean grayscale value of each vertical pixel in the two-dimensional ultrasonic grayscale image is calculated column by column, and a continuous and smooth vertical grayscale projection curve is obtained. This curve shows obvious numerical abrupt change inflection points in different structural regions. Based on the abrupt change inflection points, the vertical segmentation boundaries of the outer wall background region, the resistance wire feature region, and the fusion interface region are accurately divided. The boundary segmentation accuracy reaches the single pixel level, ensuring that the region division is non-overlapping and without omissions.

[0036] A pre-constructed mapping table of vertical coordinate intervals and unique mask identifiers is used. The table contains three fixed vertical coordinate intervals, each uniquely bound to the outer wall background mask, the resistance wire feature mask, and the fusion interface mask. All pixels in the image are traversed, and the vertical coordinates of each pixel are read and compared with the intervals in the mapping table. Pixels whose coordinates fall within the corresponding interval are matched with the corresponding mask identifier; pixels outside the interval are considered invalid regions and matched with blank mask identifiers. The matched mask identifiers are written to the pixel attribute field, completing intelligent mask annotation for all pixels in the image. This enables precise data routing for the three types of regions to independent processing paths.

[0037] Differentiated lightweight computational logic is applied to different regions to achieve optimal allocation of computing power: the outer wall background region has no defect analysis value, so only pixel downsampling and background color recognition are performed, and high-order operations such as edge detection, grayscale projection, and texture analysis are forcibly disabled, quickly outputting grayscale baseline parameters and significantly reducing the overall computational load; the resistance wire feature region only needs to identify its shape and position, perform contour extraction and output the resistance wire boundary coordinate array, disable global texture statistical operations, and focus on structural features; the fusion interface region is the core defect detection region, so all pixel downsampling and simplified operations are disabled, the original pixel information is fully preserved, and bilateral filtering, grayscale projection, and sub-pixel edge extraction are performed simultaneously to accurately output the grayscale parameters, texture parameters, and boundary distortion parameters of the fusion area, providing high-precision image data for subsequent defect feature extraction.

[0038] Specifically, the mean vertical grayscale value is calculated column by column for the two-dimensional grayscale image to generate a continuous vertical grayscale projection curve. The abrupt change points of the curve are identified by the gradient difference method to determine the vertical segmentation boundary of the three-layer region of the outer wall background, resistance wire, and fusion interface.

[0039] The device pre-stores a vertical range-mask identifier mapping table, with three sets of vertical coordinate ranges corresponding to three types of region masks; it reads the vertical coordinate values ​​of the image pixel by pixel to perform matching judgment. If the vertical coordinate of a pixel falls between the upper and lower limits of a certain range, it directly matches the corresponding area mask identifier. The vertical coordinate of a pixel is equal to the upper and lower limits of the interval, and all pixels are uniformly classified into this interval to avoid missing boundary pixels. If the vertical coordinate of a pixel exceeds the range of the three intervals, it will be matched with a blank mask and will not participate in subsequent feature calculations.

[0040] The matched mask identifiers are written into the pixel's built-in attribute field to generate three independent pixel masks, which are then distributed to three parallel processing chains for differential calculation: Outer wall background link: Perform 2x pixel downsampling, only calculate the average grayscale value of the background color as the reference parameter, permanently disable edge detection, grayscale projection, and texture solving operations, and reduce 70% of invalid computing power; Resistance wire feature chain: Only the Canny contour extraction algorithm is executed, and the coordinate array of the closed boundary of the resistance wire is output. Global texture statistics are disabled, and only the geometric shape extraction function is retained. The core link of the fusion interface: without performing any downsampling or pixel simplification operations, it fully performs bilateral filtering, vertical grayscale projection, and sub-pixel edge extraction with full high-precision calculations, thus fully preserving the subtle grayscale variation characteristics of the cold-welded thin layer.

[0041] For the outer wall background area, the image data quantization bit width of this area is reduced from 32-bit floating point number to 8-bit unsigned integer, and 3×3 kernel mean filtering is performed; For the characteristic region of the resistance wire, 16-bit fixed-point quantization is used, and edge detection based on the Sobel operator is performed; For the core region of the fusion interface, 32-bit floating-point arithmetic precision is retained, and anisotropic bilateral filtering based on a 5-pixel window radius is performed.

[0042] For the background region of the outer wall, a lightweight compression method is used to increase the downsampling interval and reduce the number of filtering iterations. By increasing the horizontal and vertical pixel sampling step size of the image, the overall pixel data volume is reduced, and the number of iteration operations of the image smoothing filtering and region segmentation algorithm is reduced, thereby reducing the computational burden on invalid regions. For the characteristic region of the resistance wire, a medium-precision computing strategy with precision threshold constraint is adopted. By appropriately compressing the image gray-scale quantization level and reducing the precision of gradient edge extraction, the periodic bright ripple contour features of the resistance wire are stably preserved and the position parameters of the upper and lower reference boundaries of the fusion section are reliably obtained. At the same time, the computing power consumption in this region is appropriately controlled to ensure the positional stability and computing timeliness of the reference feature extraction. For the core region of the fusion interface that determines the accuracy of cold welding defect identification, all lightweight compression and accuracy degradation strategies are turned off. The complete set of high-precision calculation logic, including full-precision bilateral filtering, global grayscale projection calculation, and sub-pixel edge extraction, is fully preserved. No quantization compression or detail discarding is performed on the image features of the fusion interface region. The weak ultrasonic reflection grayscale details and continuous texture features corresponding to the thin cold welding layer at the fusion interface of the electrofusion joint are fully preserved. Ultimately, a two-way balance is achieved between the overall computing efficiency of the embedded terminal and the accuracy of weak cold welding defect identification.

[0043] S3. Multi-dimensional defect feature extraction and differentiated fixed-point quantization encoding and storage The core function of this step is to extract four types of cold welding-specific features from the image that are quantifiable, calculable, and distinguishable. This transforms the hidden defect information that is not visible to the naked eye into standardized values ​​that can be recognized by the machine, solving the problems of traditional detection relying on human experience, features that cannot be quantified, and the strong subjectivity of defect judgment.

[0044] In practice, based on the high-precision image after partitioning and lightweight processing, the accurate extraction of four major categories of core features is achieved: One feature is the geometric spacing characteristic: the coordinates of the center of the resistance wire feature area skeleton and the effective edge pixel coordinates of the fusion interface are retrieved, a two-dimensional difference conversion is performed, and the pixel size is converted to the actual physical size by combining the image physical resolution parameters, generating vertical spacing sub-features in the wall thickness direction and circumferential distribution spacing sub-features respectively. This feature can accurately reflect typical structural defects of cold welding such as uneven fusion layer thickness, interface offset, and distortion of the relative position of the resistance wire and the fusion layer.

[0045] Secondly, the interface echo amplitude characteristic is generated by traversing all effective pixels in the fusion interface region, statistically analyzing multi-dimensional gray-level parameters such as gray-level mean, gray-level variance, maximum gray-level attenuation, and gray-level range, and then fusing them to generate the interface echo amplitude characteristic. Due to incomplete fusion and micro-gaps at the interface, cold welding defects cause a significant attenuation of ultrasonic reflection energy, resulting in an echo amplitude significantly lower than that of a well-fused region. This characteristic can directly quantify the degree of bonding density at the fusion interface.

[0046] Thirdly, the texture uniformity feature of the fusion zone: an 8-neighbor gray-level co-occurrence matrix is ​​constructed for the pixel matrix of the fusion interface region, and high-order texture sub-features such as texture entropy, contrast, correlation, uniformity, and roughness are extracted and fused to generate the texture uniformity feature of the fusion zone. In a well-fused region, the pixel distribution is uniform and the texture is regular; in a cold-welded defect region, the micro-gap is chaotic, the texture is disordered, and the uniformity is extremely poor. This feature can effectively distinguish between uniform cold-welded defects and local point defects.

[0047] Fourthly, the feature of continuous circumferential distribution of defects: Automatically identifies continuous pixel clusters with grayscale and texture anomalies within the fusion interface area, calculates the circumferential extension length of these abnormal pixel clusters, and combines this with the total circumferential pixel length of the fusion interface to generate a circumferential coverage ratio feature. Cold welding defects are mostly distributed in a continuous band, while holes and impurities are mostly distributed as discrete points. This feature is a core indicator for distinguishing cold welding from other defects.

[0048] To achieve machine intelligent discrimination, this invention employs a differentiated fixed-point quantization coding strategy: For the four types of features with varying data volumes, numerical ranges, and precision requirements, 8-bit, 16-bit, and 32-bit integer coding formats with different bit widths are matched respectively; the actual physical values ​​of all features are mapped to the corresponding coding intervals using a normalization algorithm, eliminating dimensional differences and achieving unified quantization coding. A pre-constructed structured storage array is used, allocating independent fixed storage address segments for each type of feature. The standardized coded feature data is stored in partitions, ensuring that feature data can be quickly retrieved, accurately matched, and cross-operated, providing standardized data support for subsequent multi-layer intelligent discrimination.

[0049] Specifically, based on the image data processed by S2 partitioning, four types of defect features with completely distinguishable physical meanings are extracted simultaneously. After all features are normalized and fixed-point encoded, they are stored in partitions, solving the industry pain point that a single geometric feature cannot distinguish multiple types of near-interface defects. Geometric spacing features: Read the coordinates of the resistance wire outline skeleton and the sub-pixel boundary coordinates of the fusion interface, calculate the coordinate difference in the vertical wall thickness direction and the circumferential direction, and combine the physical resolution of the image pixels to complete the unit conversion, generating two sets of sub-features: vertical spacing and circumferential spacing, which characterize the overall offset of the fusion interface. Interface echo amplitude characteristics: Extract the mean, variance, and extreme values ​​of grayscale of all effective pixels in the fusion region. The three sets of statistical parameters are coupled to form amplitude characteristics, which quantify the acoustic impedance difference of the fusion interface and are adapted to the weak reflection imaging characteristics of cold welding. Texture uniformity features of the fusion zone: The pixel matrix of the fusion zone is traversed by a 5×5 neighborhood window to construct a gray-level co-occurrence matrix. The higher-order statistics of contrast, entropy, and energy are extracted and fused into texture features to distinguish between uniform and intact fusion and non-uniform cold welding fusion. Circumferential continuous distribution characteristics of defects: Connected domains mark the gray-scale abnormal pixel clusters at the fusion interface, and the circumferential pixel extension length of the connected clusters is statistically analyzed. Dividing this length by the total circumferential pixel length at the fusion interface yields the circumferential coverage ratio of the 0~1 interval, which characterizes the continuous extension property of cold welding defects.

[0050] For the four types of feature numerical ranges and dimensional differences, 16-bit and 32-bit integer encoding formats with different bit widths are configured respectively. The original floating-point physical values ​​are linearly normalized and mapped to the encoding range to complete fixed-point quantization. The device memory is divided into four independent structured arrays to store the four types of encoded features. Each type of feature is assigned a dedicated fixed storage address segment. The data is stored in an orderly manner and the reading address is fast, which is suitable for the low memory operation environment of embedded terminals.

[0051] S4. Multi-feature hierarchical fusion discrimination and gradient risk classification of cold welding defects This invention employs a four-layer progressive multi-feature layered fusion discrimination mechanism to progressively screen and eliminate interfering defects, accurately identify cold welding defects, and effectively distinguish cold welding defects from point-like fusion holes, pipe wall oxidation impurities, and pipe socket misalignment defects, as shown in the attached figure. Figure 2 As shown, it includes the following steps: S401, Initial Defect Screening: Based on the characteristics of interface echo amplitude and the continuous circumferential distribution of defects, an initial screening is performed. When the interface echo amplitude significantly decreases and is less than the first preset threshold, and the circumferential coverage ratio of the defects is greater than the second preset threshold, it is determined to be a candidate area for cold welding defects. This layer mainly removes intact areas without continuous anomalies and sporadic areas with slight noise anomalies, quickly locating suspected defect areas.

[0052] S402, Position Compliance Verification: Based on geometric spacing characteristics, calculate the deviation of the fusion spacing of the candidate area from the standard fusion spacing. When the deviation exceeds the third preset threshold, it indicates that the abnormality is caused by pipe socket offset or assembly misalignment, rather than cold welding. Such interference is directly eliminated, and candidate areas that conform to the physical fusion law of electrofusion joints are retained to thoroughly distinguish the defect of incomplete socketing.

[0053] S403 Texture Compliance Verification: Based on the texture uniformity characteristics of the fusion zone, the candidate area is textured. If the texture uniformity of the area is high, the texture is regular and greater than the fourth preset threshold, it is judged as a local uniform anomaly such as point-like fusion holes or pipe wall oxidation impurities, and is removed. Only non-uniform anomaly areas with messy textures and poor uniformity are retained to accurately distinguish between local point defects and global cold welding defects.

[0054] S404. Structural continuity verification: Based on the distribution morphology of the resistance wire, the remaining candidate areas are finally verified. If the anomaly is only a local point and the overall distribution of the resistance wire is continuous without distortion, it is judged as a trace impurity or a small hole interference and is removed. If the abnormal area is accompanied by slight distortion of the resistance wire, deterioration of the texture across the entire area, or continuous band-like anomaly, it is judged as a real cold welding defect area.

[0055] S405. Confirm the cold welding defect area. After four layers of progressive verification, the area that is finally retained is determined to be the real cold welding defect area of ​​the electrofusion joint, realizing the accurate identification and distinction of various similar defects.

[0056] Based on accurate defect identification, this invention establishes a quantitative risk grading system: Optimal weighting is applied to the texture uniformity characteristics of the fusion zone and the continuous circumferential distribution characteristics of defects, and a standardized cold welding confidence score is obtained through fusion calculation. A higher score indicates a more severe cold welding defect and a higher safety risk. Two preset grading thresholds are used to achieve a three-level risk classification: a confidence score below the first threshold indicates low-risk cold welding, characterized by slight localized incomplete fusion with no risk of expansion; a score between the two thresholds indicates medium-risk cold welding, with continuous defect distribution, poor fusion quality, and potential hazards under long-term pressure operation; a score above the second threshold indicates high-risk cold welding, with widespread fusion failure, extremely poor interface bonding, and significant safety hazards such as cracking and gas leakage. The final output is a quantitative multi-level cold welding risk evaluation result, providing a precise basis for operation and maintenance rectification.

[0057] Specifically, six categories of standard reference samples are prepared in advance: defect-free intact electrofusion joints, gradient-sized pure cold welded samples, single-point fusion cavity samples, pipe wall oxide interlayer samples, socket gap defect samples, and mixed defect samples; 15 sets of gradient samples are prepared for each category to cover all working conditions. The unified process S1-S3 of this invention is used to collect four-dimensional feature data of all samples, statistically analyze the distribution range of various defect feature values, and solidify the critical values ​​of adjacent defect intervals as the first to fourth preset thresholds and two-level risk classification thresholds. The parameters are written into the read-only storage area of ​​the equipment and uniformly solidified at the factory, requiring no manual on-site adjustment, and the scheme can be stably reproduced.

[0058] The texture uniformity feature and the proportion of continuous defects in the circumference are assigned fixed weighting coefficients and summed to calculate the cold welding confidence score in the range of 0 to 100. The equipment has a built-in two-level classification threshold: confidence <35 is low risk, 35~70 is medium risk, and >70 is high risk. The classification result is used as the sole basis for subsequent detection mode switching and a standardized multi-level risk assessment report is output.

[0059] All the first, second, third, and fourth preset thresholds, as well as the two-level cold welding risk classification thresholds, were obtained through standard sample testing and have a clear and reproducible acquisition process. The specific calibration implementation method is as follows: Multiple sets of gradient standard samples were prepared. The sample system included: defect-free PE electrofusion joint samples, pure cold welding defect samples with different circumferential lengths, point-like fusion hole samples, pipe wall oxidation impurity samples, and pipe socket defect samples. The number of each type of sample was equal. The cold welding samples were set with multiple continuous circumferential length gradients to cover different working conditions from slight cold welding to large-scale cold welding.

[0060] Using the complete signal acquisition and feature extraction process S1 to S3 of this invention, the interface echo amplitude characteristics, circumferential continuous distribution characteristics of defects, geometric spacing characteristics, and texture uniformity characteristics of the fusion zone corresponding to all standard samples are collected in a unified manner, and the feature datasets of various types of samples are summarized.

[0061] First preset threshold: Statistically analyze the amplitude characteristic range of the interface echo between the intact fusion region and the cold welding defect region, and take the critical value at the boundary between the two types of sample intervals as the first preset threshold; Physical meaning: Distinguish between high amplitude echo of intact fusion and weak amplitude echo of cold welding.

[0062] The second preset threshold is determined by statistically analyzing the circumferential continuous distribution characteristics of discrete point defects and continuous cold welding defects, and selecting the intermediate critical value between the maximum coverage ratio of discrete defects and the minimum coverage ratio of continuous cold welding defects as the second preset threshold; this is used to screen abnormal areas with continuous distribution characteristics.

[0063] The third preset threshold is to collect the geometric spacing characteristic offset of multiple qualified joints and defective joints with incomplete insertion, and use the maximum allowable offset of qualified joints as the third preset threshold; this is used to eliminate the overall interface offset abnormality caused by incomplete insertion of pipes.

[0064] The fourth preset threshold is to compare the texture uniformity distribution of the fusion zone in the well-fused area and the cold-welded area, and select the lower limit of the texture uniformity of the well-fused sample as the fourth preset threshold; to distinguish between the qualified interface with uniform melting and the cold-welded interface with disordered melting.

[0065] Two-level risk classification thresholds: The confidence scores of cold welding of all cold welding samples are summarized and classified according to the circumferential extension length of defects and the degree of incomplete fusion. The confidence values ​​corresponding to the classification boundary are set as the first classification threshold and the second classification threshold in sequence, and always satisfy: the first classification threshold < the second classification threshold.

[0066] S5, Adaptive Dual-Mode Secondary Discrimination and Encrypted Traceability Report Generation The core function of this step is to adaptively switch the detection mode according to the risk level, balancing the efficiency of batch inspections with the accuracy of high-risk defect detection, while realizing full-process data encryption archiving and traceability management, adapting to the needs of large-scale pipeline network operation and maintenance inspection.

[0067] In practice, the multi-level assessment results of cold welding risk output by S4 are read, and the working mode is intelligently switched through preset branch control logic: When a low-risk cold welding defect is identified, the system automatically switches to the rapid inspection mode, which only retrieves geometric spacing features for quick verification. After confirming that there are no structural fusion abnormalities, the system drives the scanning mechanism to move quickly according to the preset standard displacement step size and directly enters the next electrofusion joint inspection, which greatly improves the efficiency of batch pipeline inspection and avoids minor abnormalities from occupying high-precision computing power.

[0068] When a cold welding defect is identified as medium-risk or high-risk, the system automatically switches to the precise re-inspection mode, locks the current joint spatial position, restarts the 360° full-coverage circumferential scan, and acquires high-precision ultrasonic signals a second time. At the same time, it calls the built-in bilateral filtering and gray-level co-occurrence matrix high-order statistical solution full feature extraction algorithm to perform comprehensive and refined feature analysis of the defect area, complete the second precise judgment, minimize the random error of the initial judgment, and prevent high-risk defects from being missed or under-judged.

[0069] After the second assessment, the system automatically compares the initial risk level with the reassessment result, using the second accurate assessment result as the final basis to update the cold welding risk assessment level of the current electrofusion joint, ensuring rigorous and accurate test results. After verification, the system drives the scanning mechanism to move and begin testing the next joint.

[0070] After the entire testing process is completed, report generation and data archiving are performed: the original one-dimensional radio frequency time-domain ultrasonic signal, two-dimensional ultrasonic grayscale image, four types of quantitative feature data, preliminary risk assessment results, and secondary verification results are all linked and bound in all dimensions to form a complete closed loop of testing data; an asymmetric encryption algorithm is used to encrypt the bound set of testing data to prevent data tampering and loss, and a standardized cold welding defect identification report is generated; finally, the encrypted report is persistently stored, and an index directory containing the testing timestamp, on-site geographic location information, and unique number of the electrofusion joint is established to achieve accurate traceability, verifiability, and auditability of the testing data of each joint, meeting the requirements of digital, standardized, and traceable operation and maintenance for gas pipeline safety testing.

[0071] This embodiment adopts an intelligent detection process of "screening first, then diagnosis," and adaptively switches between rapid inspection and precise re-inspection modes based on the cold welding risk assessment results. The specific logic is as follows: First, the system defaults to a fast inspection mode. In this mode, the embedded unit performs high-bit-width compression on the outer wall background area, simplifies the texture feature extraction process for the core area of ​​the fusion interface (e.g., only calculating local variance, omitting high-order operations on the gray-level co-occurrence matrix), and mainly relies on geometric spacing features for preliminary screening. This mode has a low computational load and can achieve a scanning speed of no less than 10 joints per second.

[0072] During rapid inspection, once the system determines that there is an anomaly at the current joint based on the currently extracted feature vector and outputs a multi-level cold welding risk assessment result, the system immediately activates the mode adaptive switching mechanism: No risk / Low risk: The system determines that it is normal or has only slight oxidation inclusions, maintains the rapid inspection mode, records the results and immediately moves on to the next joint for inspection, without verification.

[0073] Medium / High Risk: If the system determines that there is a suspected cold weld or improper socket insertion defect, it will immediately and automatically trigger a precise re-inspection mode. At this time, the system locks the current joint position and, based on the historical cached data or the data obtained from the re-scan, uses a full high-precision algorithm (including high-precision bilateral filtering, full-dimensional gray-level co-occurrence matrix calculation, and circumferential continuity analysis) within the core ROI of the fusion interface to perform refined feature re-extraction and secondary discrimination of the suspected defect area, in order to eliminate false alarms and confirm the severity of the defect.

[0074] Regardless of the mode, the system automatically generates an encrypted inspection report conforming to industry standards after the on-site inspection process is completed (or in real time), based on the final confirmed defect type and cold weld severity classification results. All raw RF data, B-scan images, feature parameters, and report files are stored in a local encrypted database. The entire process requires no external network support, realizing a fully automated closed loop from scanning, intelligent identification, risk classification to report output for gas PE pipeline electrofusion joints.

[0075] Implementation effect This embodiment effectively solves the technical problems of traditional PE electrofusion ultrasonic testing, such as high noise interference, low defect identification, difficulty in distinguishing similar defects, lack of quantified risk levels, and the inability to balance detection efficiency and accuracy, through a complete technical solution including adaptive noise reduction, partitioned lightweight imaging, multi-feature quantification encoding, four-layer progressive discrimination, risk gradient classification, and dual-mode adaptive re-inspection. This invention can accurately identify minute and hidden cold welding defects, effectively avoiding missed or false detections. It is widely applicable to batch on-site inspections of electrofusion joints in municipal gas PE pipelines and precise detection of key hidden dangers, possessing advantages such as high detection accuracy, strong anti-interference ability, high degree of intelligence, and data traceability.

[0076] The above provides a detailed description of an ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines, as provided in this application. Those skilled in the art will recognize that the algorithm steps of the various examples described in conjunction with the embodiments disclosed in this invention can be implemented using electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this invention. The above description of the embodiments is merely for the purpose of helping to understand the method and its core ideas of this application. It should be noted that those skilled in the art can make several improvements and modifications to this application without departing from the principles of this application, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. An ultrasonic identification method for cold welding defects in electrofusion joints of gas PE pipelines, characterized in that, include: A circumferential scan was performed on the electrofusion joint of the gas PE pipeline to obtain the one-dimensional radio frequency time-domain ultrasonic signal of the electrofusion joint. The one-dimensional radio frequency time-domain ultrasonic signal was then subjected to adaptive noise reduction processing to suppress the interference of polyethylene material scattering noise, resistance wire metal reflection clutter and probe coupling fluctuation noise. A two-dimensional ultrasonic grayscale image is generated based on the denoised one-dimensional radio frequency time-domain ultrasonic signal, and the two-dimensional ultrasonic grayscale image is divided into the outer wall background region, the resistance wire feature region and the fusion interface region, and then a differentiated lightweight processing is performed on the region. From the lightweight processed two-dimensional ultrasonic grayscale image, geometric spacing features, interface echo amplitude features, fusion zone texture uniformity features, and defect circumferential continuous distribution features are extracted, and the extracted features are differentiated fixed-point quantization encoding and storage. A multi-feature hierarchical fusion discrimination mechanism is adopted to identify cold welding defects and distinguish them from point-like fusion holes, pipe wall oxidation impurities and pipe socket defects. Based on the circumferential distribution ratio of abnormal areas at the fusion interface and the degree of feature offset, the gradient risk classification of cold welding defects is carried out, and the multi-level evaluation results of cold welding risk are output. Based on the results of the multi-level risk assessment of cold welding, the system adaptively switches between rapid inspection and precise re-inspection modes to perform secondary judgment and generate a cold welding defect identification report.

2. The method according to claim 1, characterized in that, The step of performing a circumferential scan of the electrofusion joint of the gas PE pipeline to obtain the one-dimensional radio frequency time-domain ultrasonic signal of the electrofusion joint, and performing adaptive noise reduction processing on the one-dimensional radio frequency time-domain ultrasonic signal, includes: The 0° longitudinal wave linear phased array probe is driven to fit against the outer wall surface of the electrofusion joint and performs a 360° full-coverage circumferential scan along the circumference of the electrofusion joint at a constant angular velocity. During the scan, ultrasonic beams are continuously and directionally emitted into the electrofusion joint and ultrasonic simulated echo signals reflected by the medium inside the electrofusion joint are received. The received ultrasonic analog echo signal is converted from analog to digital by the multiplexing channel in the phased array instrument to generate a one-dimensional radio frequency time domain ultrasonic signal sequence. The one-dimensional radio frequency time domain ultrasonic signal sequence contains the original time domain information of polyethylene material scattering, reflection of the built-in resistance wire of the electrofusion joint and reflection of the fusion interface. Based on the diameter, pitch, and initial phase information of the built-in resistance wire in the electrofusion joint, a resistance wire position mask is formed; Outside the area covered by the resistive wire position mask, a sliding median filter is applied to the one-dimensional radio frequency time-domain ultrasonic signal sequence along the circumferential scanning path to obtain a dynamic noise substrate, which is used to reflect the inherent scattering characteristics of polyethylene (PE) material. The dynamic noise floor is subtracted point by point from the one-dimensional radio frequency time-domain ultrasound signal sequence to obtain the differential signal, and the wavelet soft threshold function is used to filter the obtained differential signal to remove global scattering noise. Output a noise-reduced one-dimensional radio frequency time-domain ultrasonic signal.

3. The method according to claim 1, characterized in that, The generation of a two-dimensional ultrasound grayscale image based on the denoised one-dimensional radio frequency time-domain ultrasound signal includes: The beamforming method is used to perform coherent superposition reconstruction of the noise-reduced one-dimensional radio frequency time-domain ultrasonic signal. The beamforming method performs phase compensation and energy accumulation on radio frequency time-domain ultrasonic signals at different propagation depths, and coherently superimposes the acoustic beam energy at the same depth position to generate a two-dimensional ultrasonic grayscale image. The two-dimensional ultrasonic grayscale image is used to characterize the internal cross-sectional structure of the electrofusion joint. In a two-dimensional ultrasonic grayscale image, the horizontal axis coordinate maps to the circumferential scanning position, and the vertical axis coordinate maps to the propagation depth of the ultrasonic beam in the direction of the wall thickness of the electrofusion joint. The grayscale value of the pixel in the two-dimensional ultrasonic grayscale image is positively correlated with the acoustic impedance difference and ultrasonic reflection coefficient of the medium at the corresponding position.

4. The method according to claim 1, characterized in that, The process of dividing the two-dimensional ultrasonic grayscale image into an outer wall background region, a resistance wire feature region, and a fusion interface region, and performing differentiated lightweight processing for each region, includes: The mean vertical grayscale value is calculated column by column for the denoised two-dimensional ultrasound grayscale image to obtain the vertical grayscale projection curve, and the vertical segmentation boundary of the region is determined based on the numerical abrupt change point of the vertical grayscale projection curve. Pixel mask annotation is performed on the denoised two-dimensional ultrasonic grayscale image based on the longitudinal segmentation boundary of the region, generating three sets of pixel masks. The three sets of pixel masks correspond to the outer wall background region, the resistance wire feature region, and the fusion interface region, respectively. Based on the pixel masks, the data of different regions are diverted to different computing links. For the outer wall background area, perform pixel downsampling and background color recognition, and disable edge detection, grayscale projection and texture parsing operations, and output the grayscale reference parameters of the area; For the resistance wire feature region, perform contour extraction, output the resistance wire boundary coordinate array, and disable global texture statistics operation; For the fusion interface region, downsampling and pixel simplification operations are disabled, and bilateral filtering, grayscale projection and edge extraction are performed. Full operations are performed on the pixel data in the region, and grayscale, texture and boundary parameters are output.

5. The method according to claim 4, characterized in that, The pixel mask annotation is performed on the denoised two-dimensional ultrasonic grayscale image based on the longitudinal segmentation boundary of the region, generating three sets of pixel masks, including: A mapping storage table of vertical intervals and mask identifiers is pre-established. The mapping storage table records three sets of vertical coordinate intervals, and each set of vertical coordinate intervals is bound to an independent mask identifier. After reading the vertical coordinate values ​​corresponding to the pixels of the denoised two-dimensional ultrasound grayscale image, compare them with the vertical coordinate interval values ​​recorded in the mapping storage table. When the vertical coordinate value corresponding to a pixel is between the upper and lower limits of the vertical coordinate interval, the mask identifier bound to the vertical coordinate interval is retrieved and used as the mask identifier for matching the current pixel. When the vertical coordinate value corresponding to a pixel is outside the vertical coordinate range, it does not belong to the outer wall background area, the resistance wire feature area, or the core area of ​​the fusion interface. The preset blank mask identifier is called as the mask identifier for the current pixel matching. Write the retrieved mask identifier into the pixel attribute field corresponding to the current pixel to complete the single-pixel mask identifier matching operation.

6. The method according to claim 1, characterized in that, The extraction of geometric spacing features, interface echo amplitude features, fusion zone texture uniformity features, and defect circumferential continuous distribution features, followed by differential fixed-point quantization encoding and storage of the extracted features, includes: The difference conversion operation is performed on the skeleton coordinates of the characteristic region of the resistance wire and the effective pixel coordinates of the fusion interface, and the numerical conversion is performed in combination with the image resolution parameters to generate vertical spacing sub-features and circumferential spacing sub-features, which together constitute the geometric spacing features. Extract the grayscale statistical parameters of the effective pixels in the fusion interface region and combine them to generate interface echo amplitude features; The neighborhood gray-level co-occurrence matrix is ​​solved for the pixel matrix of the fusion interface region, and multiple sets of texture sub-features are extracted from the solved gray-level co-occurrence matrix to generate texture uniformity features of the fusion area. The circumferential extension length of the continuously distributed abnormal pixel clusters within the fusion interface region is statistically analyzed, and the ratio is converted based on the total length of the circumferential pixels of the fusion interface to generate the circumferential coverage ratio sub-feature, which constitutes the circumferential continuous distribution feature of the defect. For different types of features, an integer encoding format with an appropriate bit width is selected, and the actual physical value of each feature is normalized and mapped to the encoding value range corresponding to the integer encoding format to achieve fixed-point quantization encoding. The feature data generated by the encoding process is written into a pre-configured structured array, and storage address segments are allocated for various types of feature data in the structured array for storage.

7. The method according to claim 1, characterized in that, The multi-feature hierarchical fusion discrimination mechanism performs a four-layer progressive screening of candidate regions to distinguish defect types, including: Based on the interface echo amplitude characteristics and the circumferential continuous distribution characteristics of defects, a preliminary screening of cold welding defects is performed. When the interface echo amplitude characteristics are less than the first preset threshold and the circumferential continuous distribution characteristics of defects are greater than the second preset threshold, they are determined to be candidate areas for cold welding defects. Based on the geometric spacing characteristics, the position compliance of the candidate cold welding defect areas after the initial screening is checked, and the deviation value between the geometric spacing characteristics and the preset standard spacing is calculated. When the deviation value exceeds the third preset threshold, it is removed, and the candidate cold welding defect areas whose positions conform to the physical laws of the fusion interface are retained. Based on the texture uniformity characteristics of the fusion zone, the texture compliance of the candidate regions of cold welding defects retained after position verification is checked. When the texture uniformity characteristics are greater than the fourth preset threshold, they are removed, and candidate regions of cold welding defects with non-uniform texture characteristics are retained. Based on the distribution characteristics of the resistance wire feature region, the structural continuity of the cold welding defect candidate region retained after texture verification is checked. When the cold welding defect candidate region shows local point-like anomalies and the resistance wire distribution remains continuous, it is removed, and the cold welding defect candidate region with resistance wire distribution distortion or global texture deterioration is retained. The candidate areas for cold welding defects retained after four layers of verification are determined as cold welding defect areas.

8. The method according to claim 1, characterized in that, The method of classifying cold welding defects into gradient risks based on the circumferential distribution ratio and feature offset of abnormal regions at the fusion interface outputs multi-level evaluation results of cold welding risks, including: The texture uniformity characteristics of the fusion zone and the circumferential continuous distribution characteristics of defects are weighted and calculated to obtain the cold welding confidence score. Based on the cold welding confidence score, the gradient risk classification of cold welding defects is carried out. When the confidence score of cold welding is less than the first classification threshold, it is judged as a low-risk cold welding defect; When the cold welding confidence score is greater than the first level threshold but less than the second level threshold, it is judged as a medium-risk cold welding defect. When the cold welding confidence score is greater than the second-level threshold, it is judged as a high-risk cold welding defect; Wherein, the second grading threshold is greater than the first grading threshold.

9. The method according to claim 1, characterized in that, The process involves adaptively switching between rapid inspection and precise re-inspection modes based on the multi-level risk assessment results of cold welding, and performing secondary judgment, including: Read the multi-level risk assessment results of cold welding and match the preset branch control logic to perform the switching operation between rapid inspection and precise re-inspection working modes; When the multi-level assessment result of cold welding risk is low-risk cold welding defect, switch to the rapid inspection mode, retrieve the geometric spacing features for preliminary screening, drive the scanning mechanism to complete the displacement according to the preset displacement step size, and switch to the next electrofusion joint to carry out cold welding defect identification. When the multi-level evaluation result of cold welding risk is medium-risk cold welding defect or high-risk cold welding defect, switch to the precise re-inspection mode, re-execute the circumferential scanning for the current spatial position of the electrofusion joint, obtain the ultrasonic signal, and call the full feature extraction algorithm for secondary discrimination. The full feature extraction algorithm has built-in bilateral filtering operation and gray-level co-occurrence matrix high-order statistical quantity solution operation. After completing the secondary discrimination, the output result of the secondary discrimination is compared with the multi-level evaluation result of cold welding risk. If the output result of the secondary discrimination is a low-risk cold welding defect, the cold welding risk evaluation result of the current electrofusion joint is updated to a low-risk cold welding defect, and the scanning mechanism is driven to complete the displacement according to the preset displacement step size, and switch to the next electrofusion joint to carry out cold welding defect identification.

10. The method according to claim 9, characterized in that, The generation of the cold welding defect identification report includes: The one-dimensional radio frequency time-domain ultrasound signal, the two-dimensional ultrasound grayscale image, the extracted features, the multi-level evaluation results of cold welding risk and the secondary discrimination results are associated and bound together. Asymmetric encryption algorithm is used to encrypt the associated data to generate a cold welding defect identification report; The encrypted cold welding defect identification report is stored, and an index directory containing timestamps, geolocation information, and electrofusion joint numbers is created.

Images