A stainless steel composite straight seam welded pipe weld defect intelligent detection method

Abstract

The application discloses a kind of intelligent detection methods for weld defects of stainless steel composite straight welded pipe, belong to nondestructive intelligent detection technical field.The method includes: establishing longitudinal space coordinate axis mapped with the length direction of weld;Synchronous acquisition of phased array ultrasonic fan-shaped scanning reflected echo signal and pulse eddy current induced eddy current attenuation curve, respectively extract interface reflection peak amplitude and decay time constant;Equivalent thickness of coating is calculated using decay time constant, and after comparing with nominal thickness, the receiving gain of phased array ultrasonic is dynamically adjusted, and the acoustic characteristics after compensation are output;The compensated acoustic characteristics and electromagnetic attenuation characteristics are input into the fusion judgment model to identify the defect category and spatial position.The application calibrates ultrasonic gain in real time through pulse eddy current, eliminates heterogeneous interface acoustic impedance mismatch interference, combines multimodal characteristics and support vector machine classification, realizes weld defect detection in full thickness range, reduces false defect alarm rate, and is suitable for online automatic detection.

Description

Technical Field

[0001] This application belongs to the field of non-destructive intelligent testing technology, specifically relating to an intelligent detection method for weld defects in stainless steel composite straight seam welded pipes. Background Technology

[0002] Stainless steel composite straight seam welded pipe is made of carbon steel base layer and stainless steel cladding layer, which are composite rolled and welded together. Its weld area combines the strength of carbon steel and the corrosion resistance of stainless steel, and it is widely used in oil, natural gas and pressure vessel fields. The quality of the weld directly determines the operational safety and service life of the pipeline. Therefore, accurate and efficient detection of internal and surface defects of the weld on the production line is a key link in quality control.

[0003] Currently, non-destructive testing (NDT) for stainless steel composite welds mainly employs ultrasonic testing and eddy current testing. Ultrasonic testing can penetrate relatively thick walls, but it is sensitive to interfaces between dissimilar materials: when sound waves pass through the composite interface of carbon steel and stainless steel, the difference in acoustic impedance and microstructure between the two materials generates strong interface reflection waves and background clutter, easily drowning out the signal characteristics of real, minute defects, leading to missed detections or false alarms. Eddy current testing is highly sensitive to surface and near-surface defects, but is limited by the skin effect, making it difficult to effectively penetrate the deep regions of thick-walled welds and reliably detect deep cracks, lack of fusion, and other defects on the carbon steel base layer. Furthermore, the thickness of the stainless steel cladding may fluctuate in actual production; this geometric variation further alters the interface acoustic reflection intensity. Existing testing methods lack real-time sensing and compensation mechanisms for cladding thickness changes, failing to distinguish between material geometric fluctuations and real defects, thus affecting the stability and reliability of the test results.

[0004] Therefore, how to overcome the limitations of a single detection method at heterogeneous interfaces, achieve full thickness and interference-resistant detection from the surface of the coating to the depth of the base layer, and automatically eliminate the interference of false defects caused by coating thickness fluctuations is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent detection method for weld defects in stainless steel composite straight seam welded pipes, which can effectively solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] A smart detection method for weld defects in stainless steel composite straight seam welded pipes includes the following specific steps:

[0008] S1: Establish a longitudinal spatial coordinate axis. Move the detection frame along the weld seam, use the displacement encoder to output pulse signals and accumulate them to establish a spatial coordinate axis that maps to the weld seam length direction.

[0009] S2: Acoustic data is acquired synchronously. The phased array ultrasonic probe performs a sector scan on the weld, records the reflected echo signal, and extracts the peak amplitude of the interface reflected wave.

[0010] S3: Synchronously extract electromagnetic attenuation characteristics. Apply excitation pulses to the pulsed eddy current sensor and record the induced eddy current attenuation curve to extract the attenuation time constant characterizing the coating thickness.

[0011] S4: Perform dynamic gain compensation, align acoustic and electromagnetic data with the spatial coordinate axes; calculate the equivalent thickness of the coating using the attenuation time constant, compare it with the nominal thickness, dynamically adjust the receiving gain of the phased array ultrasound, and output the compensated acoustic characteristics.

[0012] S5: Intelligent recognition, which inputs the compensated acoustic features and electromagnetic attenuation features into the fusion judgment model to identify the defect type and spatial location.

[0013] Furthermore, the establishment of the longitudinal spatial coordinate axis in step S1 further includes: outputting dual-channel orthogonal pulse signals with a phase difference of 90 degrees through a high-resolution photoelectric encoder; the orthogonal decoding logic module inside the data processing unit performs edge detection and 4-fold frequency multiplication on the input A and B phase pulse signals, thereby effectively increasing the frequency of the original pulse sequence by 4 times.

[0014] During the detection process, the output increment of the displacement encoder is compared with the motion state sensed by the accelerometer integrated on the detection frame in real time to detect whether the drive wheel slips on the surface of the pipe. When pulse loss or discontinuity is detected, an early warning signal is triggered and data recording of the current scanning area is paused.

[0015] Further, the extraction of the peak amplitude of the interface reflected wave in step S2 specifically includes: based on the nominal wall thickness of the stainless steel composite straight seam welded pipe to be tested, setting a dynamic following time window in the acoustic scanning data corresponding to the interface position of the stainless steel cladding and the carbon steel base layer.

[0016] Within the dynamic tracking time window, the peak amplitude of the interface reflected wave is located using a peak search algorithm and recorded as the original acquisition amplitude. ;

[0017] Based on the current beam deflection angle and the material's attenuation characteristics for sound waves, the original acquired amplitude is... The correction process is performed, and the correction formula is as follows: ,in The deflection angle of the synthesized sound beam. The sound attenuation coefficient is... Given the one-way sound path length, the corrected interface reflection wave amplitude is obtained. .

[0018] Furthermore, step S2 also includes extracting a multidimensional acoustic feature vector characterizing the defect properties from the reflected echo signal. The multidimensional acoustic feature vector includes: the full width at half maximum (FWHM) of the echo envelope, used to quantify the duration of the defect echo in the time domain.

[0019] Spectral centroid displacement, by comparing the spectral centroid shift of the defect echo and the interface reflected wave, characterizes the selective attenuation or scattering effect of the defect on the frequency components of the sound wave.

[0020] And the energy centroid depth of the signal, by calculating the weighted average position of the echo energy in the time domain, the equivalent depth position of the defect in the wall thickness direction is determined.

[0021] Further, the extraction of the attenuation time constant in step S3 specifically includes: assuming the voltage signal corresponding to the acquired original voltage attenuation curve is... Taking the natural logarithm of the original voltage decay curve, the logarithmized signal is expressed as follows: ;

[0022] The plateauing region in the latter part of the logarithmically transformed curve is selected as the fitting interval. The least squares method is used to perform linear fitting on the data within this interval, and the fitting function is of the form: ,in The time variable is calculated from the moment the self-excitation pulse is turned off. This is the initial voltage amplitude constant term obtained from the fitting;

[0023] By solving the slope of the linear fit The decay time constant of the induced eddy current was calculated.

[0024] Principal component analysis was used to perform feature dimensionality reduction on the original attenuation curve. Attenuation curves of standard test blocks under different coating thicknesses and different lift-off distances were collected in advance to construct a sample matrix. The covariance matrix and its eigenvectors were calculated. The first eigenvector corresponding to the largest eigenvalue was selected as the principal component projection direction. The attenuation curve vector collected in real time was projected onto this principal component direction to obtain the first principal component component.

[0025] Furthermore, the real-time calculation of the equivalent thickness of the stainless steel cladding in step S4 specifically includes: pre-selecting stepped test blocks with the same material grade and cladding thickness specifications as the pipe to be tested, performing pulsed eddy current detection in each known thickness region, and recording the corresponding decay time constant. A multinomial regression method was used to establish the mapping relationship between the equivalent thickness of the coating and the decay time constant. The mapping model is as follows: ,in For the equivalent thickness of the coating, These are polynomial weighting coefficients. The order of the polynomial;

[0026] During online detection, the decay time constant will be extracted in real time. Substituting into the mapping model, the equivalent thickness of the overlay at the current spatial location is calculated. .

[0027] Furthermore, the fusion judgment model in step S5 adopts a multi-classifier architecture based on support vector machine. The feature vector input to the model has a total of 12 dimensions, specifically including: the amplitude value of the heterogeneous interface reflection wave after gain correction in step S4.

[0028] Pulse eddy current decay time constant The root mean square value of the residuals of the pulsed eddy current attenuation curve fitting, and the first principal component extracted by principal component analysis; the full width at half maximum (FWHM) of the acoustic echo envelope, the centroid displacement of the acoustic spectrum, and the energy centroid depth of the acoustic signal.

[0029] The root mean square energy value of the ultrasound A-scan signal in the region below the heterogeneous interface; the peak phase polarity characteristics of the ultrasound A-scan signal in the region below the interface; the spatial gradient of the amplitude of the interface reflection wave between adjacent scan positions; and the attenuation time constant between adjacent scan positions. The spatial gradient of the current scan position; and the location code of the weld area to which the current scan position belongs.

[0030] Furthermore, the intelligent identification and positioning in step S5 further includes: the offline training process of the fusion judgment model: collecting samples from the production line including surface cracks on the coating side, deep cracks on the base layer side, weld non-fusion, weld slag inclusions, local thinning of the coating, and defect-free normal weld areas, and having professionals perform truth value labeling through destructive testing or radiographic re-examination.

[0031] Radial basis functions are selected. As the kernel function of the support vector machine, where and They represent the first The and the first The feature vectors of each training sample The kernel function width parameter is used; multiple binary sub-classifiers are constructed using a one-to-many strategy, and the kernel function width parameter is determined through cross-validation. With penalty coefficient Online reasoning and defect classification: For each frame's normalized 12-dimensional feature vector The result is simultaneously fed into multiple trained binary sub-classifiers, with each sub-classifier outputting a decision function value. ,in For the set of support vectors, For the Lagrange multipliers corresponding to the support vectors, For the class labels of support vectors, For support vector features, For bias terms;

[0032] The class corresponding to the subclassifier with the largest and positive decision function value is selected as the determination result for the current scanning position; where, when the decay time constant... When a shift occurs and the amplitude of the ultrasonic interface reflection wave increases synchronously, it is determined to be a local thinning of the coating.

[0033] When the pulsed eddy current characteristics are stable but the ultrasonic signal shows high-energy diffraction waves below the heterogeneous interface, it is determined to be a deep crack or lack of fusion on the carbon steel base side; when the full width at half maximum (FWHM) of the ultrasonic echo envelope increases, the spectral centroid shifts to lower frequencies, and the depth of the energy centroid is within the range of the coating thickness, it is determined to be a surface crack or inclusion on the coating side.

[0034] Furthermore, the detection frame adopts an adaptive centering structure, with multiple sets of elastic guide wheels distributed at angular intervals straddling both sides of the weld, so that the lift-off distance between the phased array ultrasonic probe and the pulsed eddy current sensor relative to the weld surface is maintained within a preset range.

[0035] The detection frame is also equipped with an automatic spray coupling system that dynamically adjusts the acoustic coupling agent flow rate according to the detection speed. The data processing unit integrates a general-purpose processor and an FPGA hardware acceleration module. The FPGA hardware acceleration module obtains the gain compensation value by looking up the equivalent thickness deviation calculated in step S4, and writes the control word into the digital variable gain amplifier register inside the phased array controller through the serial peripheral interface to perform real-time scaling of the echo signal amplitude.

[0036] Furthermore, the fusion judgment model has online learning capabilities. The system automatically records the defect data of each judgment and its corresponding multimodal raw signal. After the inspection personnel complete the manual review and enter the results, the online learning module compares the differences between the manual judgment and the system judgment, and uses an incremental learning strategy to automatically update the support vectors and corresponding weight coefficients of the support vector machine classifier.

[0037] The method includes an automatic response mechanism for abnormal situations: during the coordinate establishment phase, if the displacement encoder loses pulses due to slippage, the system senses the motion status through the built-in accelerometer and triggers an early warning in real time, pausing data recording until the coordinates are recalibrated.

[0038] During data acquisition, if insufficient water supply from the sprinkler system causes acoustic coupling failure, the processing center will identify the overall drop in background wave amplitude and automatically mark it as an invalid detection area in the record. Once coupling is restored, a retest will be prompted.

[0039] In summary, this application includes at least one of the following beneficial technical effects:

[0040] 1. This invention fundamentally solves the acoustic impedance mismatch problem at the interface of stainless steel composite materials by performing real-time physical calibration of phased array ultrasonic signals using pulsed eddy current signals. The equivalent coating thickness calculated using the principle of electromagnetic induction provides a dynamic reference for acoustic gain, thereby significantly reducing the false defect alarm rate caused by uneven coating thickness. When identifying microcracks below a predetermined size, the signal-to-noise ratio is greatly improved, ensuring high reliability of the detection results.

[0041] 2. This invention combines the deep penetration capability of phased array ultrasound with the near-surface high sensitivity of pulsed eddy current. The phased array acoustic beam covers deep incomplete fusion and inclusion defects on the substrate side through fan-shaped scanning, while the pulsed eddy current effectively captures micro-cracks on the coating surface that are significantly affected by the skin effect. The high degree of integration of the two in spatial coordinates ensures that the entire weld area from the inner wall to the outer wall of the pipe is under close monitoring, filling the detection blind spot at the interface of heterogeneous materials using traditional single methods.

[0042] 3. By establishing a multimodal fusion judgment model, this invention can automatically decouple the metallurgical fluctuations of the material itself from the actual physical defect characteristics. The system no longer relies on a single amplitude threshold judgment, but instead performs intelligent classification based on a multidimensional mapping of acoustic feature vectors and electromagnetic attenuation characteristics. This not only significantly reduces the reliance on the experience of inspection personnel, but also enables the visualization of defect properties through three-dimensional viewing technology, providing accurate data support for subsequent stainless steel composite straight seam welded pipe repair processes.

[0043] 4. This invention integrates displacement encoding logic, ultrasonic array driving, and electromagnetic pulse excitation into a unified control framework, enabling synchronous data acquisition and real-time compensation calculation during continuous movement. The adaptive structure of the detection frame, combined with a high-speed hardware processing unit, allows this method to fully meet the online inspection requirements of large-scale industrial production lines. While ensuring the predetermined inspection coverage, the detection efficiency of a single pipe is significantly improved compared to traditional offline inspection, greatly optimizing the quality control process for manufacturing enterprises. Attached Figure Description

[0044] Figure 1 A schematic diagram of the overall technical solution for an intelligent detection method for weld defects in stainless steel composite straight seam welded pipes;

[0045] Figure 2 A schematic diagram illustrating the core principles of multi-source data fusion and dynamic gain compensation;

[0046] Figure 3 The flowchart shows the logic of acoustic scanning and electromagnetic induction feature acquisition based on displacement encoder synchronization.

[0047] Figure 4This is a schematic diagram of the data flow for multidimensional feature extraction and intelligent classification and recognition of defects based on a fusion judgment model;

[0048] Figure 5 This is a flowchart illustrating the logic of equivalent thickness mapping and dynamic adjustment of receiver gain to address heterogeneous interface interference. Detailed Implementation

[0049] To make the objectives, technical solutions, and advantages of this invention clearer, the following description is provided in conjunction with the appendix. Figure 1 To be continued Figure 5 The present invention will be further described in detail below with reference to specific embodiments.

[0050] In this embodiment, the specific implementation process of the intelligent detection method for weld defects in stainless steel composite straight seam welded pipes is based on the precise digital definition of the spatial location of the weld.

[0051] First, in step S1, a high-resolution longitudinal coordinate axis that strictly maps to the physical orientation of the weld is established through the coordinated operation of the displacement encoder, signal conditioning circuit, and zero-point calibration procedure. This coordinate axis serves as the spatial reference for the entire detection operation, providing a unified framework for the spatiotemporal alignment of subsequent acoustic scanning data and electromagnetic induction signals, including the following steps:

[0052] Step S101: Activate the spatial displacement sensing mechanism. After the detection operation starts, the detection frame carrying the sensing module performs linear scanning movement along the weld extension direction of the stainless steel composite straight seam welded pipe. The displacement encoder integrated at the bottom of the detection frame is then activated and begins to sense the displacement changes of the detection frame.

[0053] This displacement encoder employs a high-resolution photoelectric encoder with an internal code disk circumferential line density of no less than 3600 lines per revolution. A drive wheel, made of high-friction polyurethane material, is linked to the encoder shaft and elastically pressed against the surface of the pipe being tested. The movement of the probe frame converts linear displacement into angular displacement of the encoder code disk without slippage through the static friction between the drive wheel and the pipe surface.

[0054] Step S102: Generating and processing orthogonal pulse signals. As the detection frame moves, the displacement encoder outputs dual-channel orthogonal pulse signals in real time. These two pulse sequences, with a 90-degree phase difference, are transmitted to the data processing unit, which serves as the core processing component of the entire detection system, responsible for data acquisition, fusion, and calculation. The data processing unit integrates an orthogonal decoding logic module, which not only records the displacement by counting pulse edges but also performs a 4-fold frequency multiplication on the input A and B phase pulse signals.

[0055] Specifically, this module effectively boosts the frequency of the original pulse sequence by a factor of four by detecting all rising and falling edge state changes of the two signals in each cycle. This process transforms continuous displacement in the physical world into high-resolution digital incremental position information.

[0056] Step S103: A high-resolution longitudinal spatial coordinate axis is established. The digital displacement increments, after being processed by a fourth harmonic, are accumulated to establish a longitudinal spatial coordinate axis that is strictly mapped one-to-one with the physical length direction of the weld. The displacement measurement accuracy of this coordinate axis is controlled at the level of 0.01 mm.

[0057] This high-precision spatial coordinate system is not an isolated measurement result, but serves as a unified spatial reference, providing a fundamental coordinate framework for high-precision spatiotemporal alignment of phased array ultrasonic scanning sections and pulsed eddy current sampling point data in subsequent steps.

[0058] Step S104: Execute the absolute zero-point calibration procedure. During the coordinate axis establishment process, the system will automatically trigger and execute the zero-point calibration procedure. This procedure aims to eliminate accumulated coordinate drift during long-distance detection. Specifically, before the detection frame begins formal data acquisition, it will first perform edge-finding or marker-finding movement along the weld seam direction.

[0059] The system determines the absolute origin of the longitudinal coordinate axis by identifying physical features at the weld end, such as the starting edge of the weld reinforcement, or by identifying magnetic marks pre-placed at specific locations on the pipe end. Once the origin is established, the spatial coordinates of all subsequent acquired data are marked based on this origin.

[0060] During continuous long-distance inspection operations, the system will also repeat the calibration procedure at preset distance intervals or according to operator instructions to ensure that the cumulative error caused by factors such as potential slippage of the drive wheels is always constrained to a controllable range below the preset threshold.

[0061] Step S105: Handling abnormal states of displacement sensing. As a protective mechanism, this method also includes logic to address displacement sensing anomalies. During the detection process, if the drive wheel accidentally slips on the pipe surface, the frequency of the pulse signal output by the displacement encoder will momentarily deviate from the actual moving speed of the detection frame.

[0062] The system compares the output increment of the displacement encoder with the motion state sensed by the accelerometer integrated on the detection frame in real time. Once a pulse loss or discontinuity is detected, an early warning signal is immediately triggered. At this time, the data processing unit pauses data recording for the currently scanned area and issues a notification to the operator. Data acquisition only resumes after the coordinate integrity verification is passed, thus avoiding defect location errors caused by spatial coordinate misalignment.

[0063] In summary, step S1 establishes a precise longitudinal spatial coordinate system for the weld seam that possesses self-verification capabilities. Based on the spatial reference provided by this coordinate system, the phased array ultrasonic acoustic scanning data acquired in step S2 and the pulsed eddy current electromagnetic attenuation characteristics extracted in step S3 can be accurately mapped to the corresponding physical location of the weld seam, thereby providing the necessary spatial positioning conditions for multi-source data fusion and dynamic gain compensation in step S4.

[0064] The next step, S2, involves the phased array ultrasonic probes mounted on the probe frame simultaneously starting to operate as the frame moves along the weld seam direction, performing acoustic scanning and data acquisition on the weld seam area. The detailed implementation process for this step is as follows:

[0065] Step S201: Configure the phased array ultrasonic probe parameters. The phased array ultrasonic probe consists of 128 independent array elements. The center frequency of each element is set within a preset frequency range of 5MHz to 7.5MHz. This frequency range is selected to balance the required penetration depth of sound waves in the stainless steel cladding and carbon steel base layer, as well as the spatial resolution capability for minute defects inside the weld. The array elements are physically arranged in a linear array along a direction perpendicular to the weld.

[0066] Step S202 involves achieving fan-shaped deflection scanning of the acoustic beam. The system precisely controls the excitation timing of each array element through an electronic delay control circuit. Specifically, the excitation pulses applied to each array element have a slight relative time delay, causing the sub-sound waves emitted by each array element to interfere in space, forming a synthetic acoustic beam with a specific deflection angle and focusing depth.

[0067] By dynamically changing the delay sequence, the system drives the synthesized sound beam to perform a fan-shaped deflection scan within a predetermined angle range of -30 degrees to +30 degrees. Within a single scan cycle, the sound beam coverage area starts from the upper surface of the stainless steel cladding, passes through the heterogeneous bonding interface between the stainless steel cladding and the carbon steel base layer, and continues to extend downwards until it reaches the full thickness area of ​​the carbon steel base layer.

[0068] Step S203: Record the reflected echo signal. During propagation within the material, the acoustic impedance changes when the sound beam encounters heterogeneous interfaces, grain boundaries, or defects, resulting in a reflected echo. The phased array controller performs analog-to-digital conversion on the echo signal sensed by the receiving array elements at a sampling frequency of 100MHz, converting the original analog voltage waveform into a 12-bit resolution digital signal stream. This digital signal stream records the ultrasonic A-scan waveform data distributed along the sound path direction at each scanning angle. The A-scan waveforms from multiple adjacent scanning positions combine to form a two-dimensional acoustic S-scan map.

[0069] Step S204: Setting a dynamic time window for the heterogeneous interface. Before extracting acoustic features, the system first preprocesses the acquired acoustic scanning data. Based on the nominal wall thickness of the stainless steel composite straight seam welded pipe to be tested, the system sets a dynamic following time window in the acoustic scanning data corresponding to the interface position between the stainless steel cladding and the carbon steel base layer. The width of this time window is preset to a fixed range based on the nominal wall thickness, and slides along the sound path time axis with the scanning position, with the sound beam incident point as the zero point.

[0070] Step S205: Extract the peak amplitude of the reflected wave from the interface. Within the aforementioned dynamic time window, the system uses a peak search algorithm to locate the maximum positive or negative amplitude point within the window. This amplitude point corresponds to the peak amplitude of the reflected wave from the heterogeneous interface and is recorded as the original acquired amplitude. .

[0071] Step S206: Perform cosine correction of the sound beam incident angle. To eliminate the energy attenuation difference introduced by the oblique incidence of the phased array sound beam, the system performs cosine correction on the extracted original acquisition amplitude. A correction process is then performed. The correction is based on the current beam deflection angle and the material's attenuation characteristics for sound waves, and is calculated using the following formula:

[0072]

[0073] in: This represents the amplitude of the interface reflected wave after correction for incident angle and sound path attenuation, which is used for subsequent feature analysis and defect determination. This indicates that the raw acquisition amplitude values ​​extracted directly from the time window in step S205 are without any post-processing correction; This represents the deflection angle of the current synthesized sound beam relative to the normal direction of the phased array ultrasonic probe surface, with a value ranging from -30 degrees to +30 degrees.

[0074] This represents the acoustic attenuation coefficient of the tested material to ultrasonic waves. This coefficient is an inherent parameter of the material obtained in advance through standard test block calibration, and the unit is decibels per millimeter. This represents the unidirectional sound path length of the sound beam propagating from the surface of the phased array ultrasonic probe to the heterogeneous interface, calculated based on the geometric relationship between the sound beam deflection angle and the wall thickness.

[0075] Step S207: Extract the acoustic feature vector of the defect. In addition to the amplitude information of the interface reflected wave, the system further extracts a multi-dimensional feature vector characterizing the properties of the defect from the acoustic scanning data. This feature vector includes the following three dimensions:

[0076] The first dimension is the full width at half maximum (FWHM) of the echo envelope. This parameter is used to quantify the duration of the defect echo in the time domain, reflecting the extent of the defect along the sound beam direction.

[0077] The second dimension is the spectral centroid displacement. This parameter characterizes the selective attenuation or scattering effect of the defect on the frequency components of the sound wave by comparing the spectral centroid shift of the defect echo and the interface reflected wave.

[0078] The third dimension is the energy centroid depth of the signal. This parameter determines the equivalent depth location of the defect in the wall thickness direction by calculating the weighted average location of the echo energy in the time domain.

[0079] The three feature parameters described above characterize the scattering and absorption characteristics of the defect on the incident sound wave from the perspectives of time domain broadening, frequency domain shift, and energy distribution, respectively, providing multi-dimensional acoustic input basis for the intelligent classification of the fusion judgment model in the subsequent step S5.

[0080] In summary, step S2 completes the acquisition and feature extraction of phased array ultrasonic scanning data of the weld area. The output corrected interface reflection amplitude and multidimensional acoustic feature vector, together with the pulsed eddy current electromagnetic attenuation features extracted synchronously in step S3, constitute the original input for subsequent multi-source data fusion. Based on the precise spatial coordinate system established in step S1, the acoustic features of step S2 and the electromagnetic features of step S3 are strictly aligned in spatial position, thus providing a reliable data foundation for the dynamic calculation of the equivalent thickness of the cladding and the receiver gain compensation in step S4.

[0081] The next step is S3. While the phased array ultrasonic probe acquires acoustic scanning data in step S2, the pulsed eddy current sensor performs synchronous electromagnetic detection on the same weld coordinate region. The detailed implementation process of this step is as follows:

[0082] Step S301: Configure the pulsed eddy current sensor structure. The pulsed eddy current sensor adopts a dual-coil differential structure, including one transmitting coil and two receiving coils connected in reverse series. The transmitting coil is used to excite a transient induced current in the stainless steel cladding being measured. The two receiving coils are connected in reverse series, which allows the common-mode signals generated by environmental electromagnetic interference and lift-off fluctuations in the receiving circuit to cancel each other out, while the differential-mode signal generated by the attenuation of eddy currents inside the measured material is preserved and enhanced. This differential structure significantly improves the signal-to-noise ratio of the sensor in electromagnetic noise environments in industrial settings.

[0083] In step S302, a high-energy excitation pulse is applied, and the excitation power supply applies a high-energy pulsed current to the transmitting coil. The rise and fall times of the excitation pulse are strictly controlled to within 10 microseconds. The steep pulse edge ensures that the excitation magnetic field can quickly penetrate the stainless steel cladding and induce a transient eddy current field covering a wide frequency band. The amplitude of the pulse current is preset based on the nominal thickness of the cladding and the material conductivity of the stainless steel composite straight seam welded pipe under test to ensure that the induced eddy current can effectively cover the entire thickness range of the cladding.

[0084] Step S303: Turn off the excitation pulse and record the eddy current decay curve. When the excitation pulse is turned off, the main magnetic field generated by the transmitting coil disappears rapidly. According to the law of electromagnetic induction, the eddy current field induced inside the stainless steel cladding does not disappear immediately, but decays exponentially over time.

[0085] The secondary magnetic field generated by the eddy current during the attenuation process is captured in real time by the receiving coil and converted into a voltage signal. The sampling circuit continuously acquires this voltage signal at a sampling rate of 2MHz, recording the complete dynamic evolution process of the induced eddy current from saturation to complete disappearance, forming the original voltage attenuation curve.

[0086] Step S304: Logarithmic transformation of the attenuation curve. In the data extraction stage, the system first preprocesses the original acquired voltage attenuation curve. Since the attenuation process of induced eddy currents follows a single-exponential or multi-exponential attenuation law under ideal conditions, the system takes the natural logarithm of the voltage attenuation curve, transforming the exponential attenuation relationship into a linear relationship. Let the original voltage signal be... The signal after logarithmic processing is The conversion relationship is as follows: .

[0087] Step S305: The time constant is extracted using the least squares method. After logarithmic processing, the system selects the stable attenuation region in the latter part of the attenuation curve as the fitting interval. The starting point of this interval is determined based on the delay time after the excitation pulse is turned off to ensure that the initial transient interference has been sufficiently attenuated; the ending point is set at the effective data point before the signal amplitude drops to the noise floor. The system uses the least squares method to perform linear fitting on the logarithmic data within this interval, and the fitting function is in the form of... .

[0088] in: This represents the voltage signal value after natural logarithmic transformation in step S304, and is dimensionless. The time variable represents the time taken from the moment the self-excitation pulse is turned off, and the unit is microseconds; The constant term representing the initial voltage amplitude obtained from the fitting corresponds to the theoretical voltage value at the start of the decay process, in volts. This represents the decay time constant of the induced eddy current, expressed in microseconds.

[0089] The slope of the linear fit is obtained by using the least squares method. The system can then calculate the decay time constant. This time constant directly reflects the comprehensive physical characteristics of the stainless steel cladding, including its electrical conductivity, magnetic permeability, and geometric thickness. A thicker cladding and higher electrical conductivity result in slower eddy current decay, corresponding to a shorter time constant. The larger.

[0090] Step S306: Principal component analysis is used to extract robust features. To further improve the robustness of electromagnetic features to weak interference, the system also uses principal component analysis to perform feature dimensionality reduction and extraction on the original attenuation curve. Specifically, the system pre-collects multiple sets of attenuation curves of standard test blocks under different coating thicknesses and different lift-off distances to construct a sample matrix.

[0091] After centering the sample matrix, the covariance matrix and its eigenvalues ​​and eigenvectors are calculated. The first eigenvector corresponding to the largest eigenvalue is selected as the principal component projection direction. During online detection, the attenuation curve vector acquired in real time is projected onto this principal component direction to obtain the first principal component component. This component serves as the core electromagnetic index characterizing the coating thickness and material uniformity, and can stably reflect the actual physical state of the coating even in environments with weak electromagnetic interference or lift-off fluctuations.

[0092] In summary, step S3 completes the synchronous acquisition and feature extraction of the pulsed eddy current electromagnetic attenuation signal in the weld area. The attenuation time constant output in this step... and the first principal component component, and the corrected interface reflection amplitude extracted simultaneously with step S2. And multidimensional acoustic feature vectors, to achieve strict alignment with an axial position deviation of less than 0.1 mm under the unified spatial coordinate system established in step S1.

[0093] These two types of heterogeneous feature data are input together into step S4 to calculate the equivalent thickness of the stainless steel cladding in real time, and to perform dynamic compensation adjustment of the receiving gain of the interface reflection area in the acoustic scanning data accordingly.

[0094] The next step, S4, involves spatiotemporal alignment and fusion of the acoustic scanning section data obtained in step S2 and the electromagnetic sampling point data obtained in step S3, followed by dynamic compensation of the acoustic receiving gain based on the fusion result. The detailed implementation process of this step is as follows:

[0095] Step S401: Spatial and temporal alignment of acoustic and electromagnetic data. The system uses the longitudinal spatial coordinate axis established in step S1 as a unified reference to spatially correlate the phased array ultrasonic S-scan spectrum output in step S2 with the pulsed eddy current electromagnetic attenuation characteristics output in step S3.

[0096] Through a synchronous sampling triggering mechanism activated by a displacement encoder, the acquisition times of acoustic and electromagnetic data frames are locked by the rising edge of a pulse at the same spatial position, ensuring that the corresponding deviation between each set of acoustic scanning sections and each set of electromagnetic attenuation features in the weld axial position is strictly less than 0.1 mm. After this alignment process, heterogeneous sensor data form one-to-one corresponding data pairs in the spatial dimension, providing a precise spatial matching basis for subsequent fusion calculations.

[0097] Step S402: Establish material calibration curve. Before implementing online testing, the system pre-establishes the material calibration curve using standard test blocks. Specifically, a stepped test block with the same material grade and cladding thickness as the stainless steel composite straight seam welded pipe to be tested is selected. This test block contains at least 5 cladding areas with known precise thicknesses.

[0098] Pulse eddy current detection was performed in each region of known thickness, and the corresponding decay time constants were recorded. Using the known coating thickness as the dependent variable and the corresponding measured decay time constant... As the independent variable, a multinomial regression method is used to establish the mapping relationship between the equivalent thickness of the coating and the decay time constant. This mapping relationship is stored in the non-volatile memory of the data processing unit in the form of a lookup table or a set of multinomial coefficients for real-time retrieval during online detection.

[0099] Step S403: Calculate the equivalent thickness of the stainless steel cladding in real time. During online detection, the fusion center receives the current position decay time constant output from step S3. It then calls the material calibration curve pre-stored in step S402 to calculate the equivalent thickness of the stainless steel cladding at the current location in real time. The calculation logic for the equivalent cladding thickness is based on the following mathematical correlation model:

[0100]

[0101] in: This represents the calculated equivalent thickness of the cladding, in millimeters. This value reflects the electromagnetic equivalent geometric thickness of the stainless steel cladding at the current axial position of the weld. The induced eddy current decay time constant extracted in step S305 is expressed in microseconds and is used as the input of the model's independent variable. This indicates the electrical conductivity of the stainless steel cladding material, expressed in Siemens per meter. This parameter is implicit in the calibration curve establishment process and does not require online real-time measurement.

[0102] The magnetic permeability of the stainless steel cladding material is expressed in Henry per meter, and this parameter is also implicit in the calibration curve establishment process. Indicates the first The weighting coefficients of the order polynomial, total One coefficient, From 0 to Integer index, The order of the polynomial; This represents the highest order of the fitted polynomial, which is usually an integer between 2 and 4. The specific order is determined based on the goodness of fit when the standard test block is calibrated.

[0103] The above polynomial coefficients The data is obtained by performing least-squares polynomial fitting on the known thickness points of the standard test block in step S402. During online calculation, the system will use the real-time measured decay time constant. Substituting into the polynomial, and summing the products of each power and its corresponding coefficient, we obtain the equivalent thickness of the overlay at the current spatial location. .

[0104] Step S404: Evaluate the impact of coating thickness variation on interface reflected waves. The system will use the real-time equivalent thickness calculated in step S403. The coating thickness is compared with the nominal coating thickness of the pipe to be tested. The nominal coating thickness is the design value specified in the production process documents and is pre-entered into the data processing unit.

[0105] When detected When the thickness is less than the nominal coating thickness and the deviation exceeds the preset tolerance range, the system determines that there is local thinning or geometric distortion of the coating at the current location. In this case, the acoustic impedance difference between the stainless steel coating and the carbon steel base layer is more significant than in the normal area, resulting in an abnormal increase in the energy of the heterogeneous interface reflected wave collected in step S2.

[0106] Step S405: Dynamically adjust the acoustic receiving gain. Based on the evaluation results of step S404, the system automatically adjusts the digital receiving gain of the interface reflection area corresponding to the sound beam path in the acoustic scanning data. The adjustment logic is as follows: when the calculated equivalent thickness of the cladding layer... When the thickness decreases, the system automatically reduces the gain value of the digital gain circuit corresponding to the sound beam path in the phased array receiving channel; conversely, when the equivalent thickness recovers to near the nominal value, the gain value recovers to the reference setting value. The gain adjustment range is strictly controlled within the preset adjustment range of -12 dB to 0 dB, with an adjustment step of 1 dB.

[0107] The specific implementation of digital gain adjustment is as follows: the FPGA hardware acceleration module in the data processing unit looks up the corresponding gain compensation value in the table according to the currently calculated equivalent thickness deviation, and writes the control word into the register of the digital variable gain amplifier inside the phased array controller through the serial peripheral interface to complete the real-time scaling of the echo signal amplitude.

[0108] Step S406: Output the compensated acoustic feature data. After completing the dynamic gain compensation, the system reorganizes the corrected acoustic scan data into an S-scan spectrum with the same format as the original data. The compensated acoustic feature data includes: the amplitude of the interface reflection wave after gain correction, the full width at half maximum (FWHM) of the echo envelope of the defect acoustic feature vector extracted in step S207, the spectral centroid displacement and energy centroid depth, and the pulse eddy current decay time constant output in step S3. And principal component components.

[0109] These multidimensional feature data, as a fused comprehensive feature set, are encapsulated into a unified data structure and transmitted to the fusion judgment model in the subsequent step S5 for defect identification and classification.

[0110] In summary, step S4 completes the spatiotemporal alignment of multi-source sensor data, real-time calculation of the equivalent thickness of the cladding, and dynamic compensation of acoustic receiving gain based on the thickness measurement results. The core of this step lies in utilizing the sensitive response of pulsed eddy currents to the geometry of the cladding to correct the amplitude deviation of the phased array ultrasound caused by the interface acoustic impedance mismatch in real time, thereby providing the fusion judgment model in step S5 with pure acoustic feature input that has eliminated material geometric interference.

[0111] The multidimensional feature data after compensation processing in step S4 can more realistically reflect whether there are real physical defects such as cracks, lack of fusion, or slag inclusions inside the weld, creating favorable conditions for the accurate determination of the support vector machine classifier in the subsequent step S5.

[0112] Finally, in step S5, the gain-compensated acoustic feature data and electromagnetic attenuation feature data output from step S4 are used as input and fed into a preset fusion judgment model for intelligent defect identification and spatial localization. The detailed implementation process of this step is as follows:

[0113] Step S501: Construct the input feature vector of the fusion decision model. The fusion decision model adopts a multi-classifier architecture based on support vector machines. Before feeding the data into the classifier, the system first normalizes the comprehensive feature set passed in step S4 to eliminate the influence of differences in the dimensions of each feature parameter on model training and inference.

[0114] After normalization, the feature vector received by the model input layer has a total of 12 dimensions, all of which are derived from the feature parameters extracted in steps S2, S3, and S4. Specifically, these include: the heterogeneous interface reflected wave amplitude after gain correction in step S406, the pulsed eddy current attenuation time constant τ, the root mean square value of the residual of the pulsed eddy current attenuation curve fitting, the first principal component extracted in step S306, the full width at half maximum (FWHM) of the acoustic echo envelope, the centroid displacement of the acoustic spectrum, the energy centroid depth of the acoustic signal, the root mean square energy value of the ultrasonic A-scan signal in the region below the heterogeneous interface, the peak phase polarity characteristics of the ultrasonic A-scan signal in the region below the interface, the spatial gradient of the interface reflected wave amplitude between adjacent scanning positions, the spatial gradient of the attenuation time constant τ between adjacent scanning positions, and the position code of the weld region to which the current scanning position belongs. These 12 dimensions have been defined and extracted in the preceding steps and are summarized here as input to the fusion judgment model.

[0115] Step S502, the offline training process of the support vector machine model: before implementing online detection, the fusion judgment model needs to go through an offline training stage to establish a nonlinear mapping relationship from the input feature space to the defect category label.

[0116] Regarding the construction of training data: Stainless steel composite straight seam welded pipe samples containing different defect types and defect-free states were collected from actual production lines. The defect types covered by the samples included surface cracks on the cladding side, deep cracks on the base layer side, weld incomplete fusion, weld slag inclusions, local thinning of the cladding, and defect-free normal weld areas. All samples were truth-valued by professionals with non-destructive testing qualifications through destructive testing or radiographic re-examination.

[0117] Regarding kernel function selection: Considering the nonlinear coupling relationship between acoustic and electromagnetic features, this model selects the radial basis function as the kernel function for the support vector machine, and its mathematical expression is: ,in and They represent the first The and the first The 12-dimensional feature vector of each training sample This is the kernel function width parameter.

[0118] Regarding the loss function and training objective: For the 6-class classification problem, this model employs a "one-to-many" strategy to construct 6 binary support vector machine sub-classifiers. The training objective of each sub-classifier is to solve a constrained quadratic optimization problem, with the following objective function: The constraints are and In the formula, To separate the hyperplane normal vectors, For bias terms, As slack variables, The penalty coefficient is... The total number of training samples, For the first The category label of each sample, This is a high-dimensional feature map implicitly defined through a kernel function.

[0119] During training, the kernel width parameter is determined using cross-validation. With penalty coefficient The optimal combination is determined. After training, the support vectors and corresponding parameters of each sub-classifier are saved to the non-volatile memory of the data processing unit for use in the online inference phase.

[0120] Step S503, online inference and defect classification judgment, online detection stage, for each frame, the 12-dimensional normalized feature vector constructed in step S501 The system simultaneously feeds the result into six pre-trained binary sub-classifiers. Each sub-classifier outputs a decision function value. ,in For the set of support vectors, For the Lagrange multipliers corresponding to the support vectors, For the class labels of support vectors, These are support vector features.

[0121] The system compares the decision function values ​​of the six sub-classifiers and selects the class corresponding to the sub-classifier with the largest and positive decision function value as the determination result for the current scanning position. The specific classification logic is as follows: when the decay time constant... When a significant shift occurs and the amplitude of the ultrasonic interface reflection wave increases synchronously, it is determined to be a local thinning or geometric distortion of the coating, which is a non-defect signal. When the pulse eddy current characteristics remain stable, but the ultrasonic signal shows high-energy diffraction waves or secondary reflection waves below the heterogeneous interface, it is determined to be a deep crack or non-fusion defect on the carbon steel base side. When the full width at half maximum (FWHM) of the ultrasonic echo envelope increases significantly, the spectral centroid shifts to lower frequencies, and the depth of the energy centroid is within the coating thickness range, it is determined to be a surface crack or slag inclusion defect on the coating side.

[0122] Step S504: Defect spatial location and geometric parameter calculation. For the scanned position identified as a defect in step S503, the system further uses a three-dimensional spatial projection algorithm to calculate the precise spatial coordinates and geometric dimensions of the defect. Based on the longitudinal spatial coordinate axis established in step S1, the start and end positions of the defect in the weld length direction are determined, and the longitudinal extension length is obtained.

[0123] Based on the deflection angle of the phased array ultrasonic beam and the time position of the defect echo in the A-scan waveform, the depth position and lateral offset of the defect within the weld cross-section are calculated. By performing connected component clustering analysis on pixels identified as having the same defect at adjacent detection points, the complete geometric contour of the defect is generated.

[0124] Step S505: Generate a 3D visualization view. The spatial coordinates and geometric contour data of the defects are fed into the 3D view construction module, and volume rendering technology is used to generate a semi-transparent 3D view containing the carbon steel base layer, stainless steel cladding, and weld defects. Different types and severity of defect areas are distinguished and labeled in pseudo-color within the view: red areas indicate severe defects judged as cracks or lack of fusion, yellow areas indicate minor damage judged as slag inclusions or cladding thinning, and blue semi-transparent areas represent defect-free normal weld substrate.

[0125] Step S506, Detection frame structure and hardware system configuration: The detection frame adopts an adaptive centering structure, with four sets of elastic guide wheels distributed at 90-degree intervals straddling both sides of the weld, ensuring that the lift-off distance between the phased array ultrasonic probe and the pulsed eddy current sensor and the weld surface is constantly maintained within a preset range of 0.5 mm to 1 mm.

[0126] The detection frame is equipped with an automatic spray coupling system that dynamically adjusts the acoustic coupling agent flow rate according to the detection speed. The data processing unit integrates a quad-core general-purpose processor and an FPGA hardware acceleration module. The FPGA is responsible for high-frequency pulse signal acquisition and real-time gain compensation, while the general-purpose processor is responsible for running the support vector machine classification algorithm and 3D rendering tasks. This hardware and software collaborative architecture enables a detection speed of over 500 millimeters per second, meeting the cycle time requirements for industrial online inspection.

[0127] In summary, step S5 completes the intelligent identification, spatial positioning, and visualization of weld defects. The output defect category determination results and spatial geometric parameters can be directly used to guide subsequent repair process decisions and quality level assessments.

[0128] Furthermore, the fusion judgment model possesses online learning and self-evolution capabilities. In actual production batches, the system automatically records the defect data for each judgment and its corresponding multimodal raw signals. After the inspection personnel complete the manual review and enter the results, the online learning module compares the differences between the manual judgment and the system judgment, and automatically updates the support vectors and corresponding weight coefficients of the SVM classifier using an incremental learning strategy. In this way, the system can adapt to stainless steel composite straight seam welded pipes produced with different material ratios and welding processes, improving its recognition accuracy to a target level of over 98% after 1000 hours of operation.

[0129] To further enhance the system's automation, the method also includes an automatic response mechanism for abnormal situations. During the coordinate establishment phase, if the displacement encoder loses pulses due to slippage, the system will sense the motion state through its built-in accelerometer and trigger a real-time warning, pausing data recording until the coordinates are recalibrated. During data acquisition, if insufficient water supply from the sprinkler system causes acoustic coupling failure, the processing center will identify the overall drop in background wave amplitude and automatically mark it as an "invalid detection area" in the record. Once coupling is restored, the operator will be prompted to perform a retest.

[0130] During defect location in step S5, the system not only outputs coordinate values ​​but also generates a detailed quality diagnostic report. The report includes the equivalent diameter calculation results for each defect. (Equivalent diameter) The calculation comprehensively considers acoustic reflection energy and electromagnetic disturbance range, and incorporates the coupled effects of ultrasonic beam diffusion angle and eddy current penetration depth. Its calculation expression is as follows:

[0131]

[0132] in: The equivalent diameter of the defect, expressed in millimeters; The integral energy value of the defect echo is expressed in volts squared multiplied by microseconds. This represents the half-angle of the phased array ultrasonic beam at the depth of the defect, expressed in radians. This represents the attenuation of the electromagnetic disturbance amplitude measured by the pulsed eddy current sensor in the defect area, expressed in volts. The equivalent center frequency of the pulsed eddy current excitation is expressed in Hertz. This indicates the electrical conductivity of the stainless steel cladding material, expressed in Siemens per meter. Represents the vacuum permeability, with values ​​ranging from 1 to 10. Henry per meter; , These are the weighting coefficients for the acoustic and electromagnetic terms, respectively, determined by calibration using artificial defects of known dimensions on a standard test block.

[0133] The defect contour obtained through cluster analysis is further decomposed into three components: length, width, and depth. These three components are used to construct an envelope, thereby enabling accurate estimation of the defect volume in a 3D view.

[0134] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting in all respects.

[0135] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A method for intelligent detection of weld defects in stainless steel composite straight seam welded pipes, characterized in that, include: S1: Establish a longitudinal spatial coordinate axis. Move the detection frame along the weld seam, use the displacement encoder to output pulse signals and accumulate them to establish a spatial coordinate axis that maps to the weld seam length direction. S2: Acoustic data is acquired synchronously. The phased array ultrasonic probe performs a sector scan on the weld, records the reflected echo signal, and extracts the peak amplitude of the interface reflected wave. S3: Synchronously extract electromagnetic attenuation characteristics. Apply excitation pulses to the pulsed eddy current sensor and record the induced eddy current attenuation curve to extract the attenuation time constant characterizing the coating thickness. S4: Perform dynamic gain compensation to align acoustic and electromagnetic data with the spatial coordinate axes; The equivalent thickness of the cladding is calculated using the attenuation time constant, and after comparing it with the nominal thickness, the receiving gain of the phased array ultrasound is dynamically adjusted to output the compensated acoustic characteristics. S5: Intelligent recognition, which inputs the compensated acoustic features and electromagnetic attenuation features into the fusion judgment model to identify the defect type and spatial location.

2. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The establishment of the longitudinal spatial coordinate axis in step S1 further includes: outputting dual-channel orthogonal pulse signals with a phase difference of 90 degrees through a high-resolution photoelectric encoder; the orthogonal decoding logic module inside the data processing unit performs edge detection and 4-fold frequency multiplication on the input A and B phase pulse signals, effectively increasing the frequency of the original pulse sequence by 4 times. During the detection process, the output increment of the displacement encoder is compared with the motion state sensed by the accelerometer integrated on the detection frame in real time to detect whether the drive wheel slips on the surface of the pipe. When pulse loss or discontinuity is detected, an early warning signal is triggered and data recording of the current scanning area is paused.

3. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The extraction of the peak amplitude of the interface reflected wave in step S2 specifically includes: based on the nominal wall thickness of the stainless steel composite straight seam welded pipe to be tested, setting a dynamic following time window in the acoustic scanning data corresponding to the interface position of the stainless steel cladding and the carbon steel base layer. Within the dynamic tracking time window, the peak amplitude of the interface reflected wave is located using a peak search algorithm and recorded as the original acquisition amplitude. ; Based on the current beam deflection angle and the material's attenuation characteristics for sound waves, the original acquired amplitude is... The correction process is performed, and the correction formula is as follows: ,in The deflection angle of the synthesized sound beam. The sound attenuation coefficient is... Given the one-way sound path length, the corrected interface reflection wave amplitude is obtained. .

4. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, Step S2 further includes extracting a multidimensional acoustic feature vector characterizing the defect properties from the reflected echo signal. The multidimensional acoustic feature vector includes: the full width at half maximum (FWHM) of the echo envelope, used to quantify the duration of the defect echo in the time domain. Spectral centroid displacement, by comparing the spectral centroid shift of the defect echo and the interface reflected wave, characterizes the selective attenuation or scattering effect of the defect on the frequency components of the sound wave. And the energy centroid depth of the signal, by calculating the weighted average position of the echo energy in the time domain, determines the equivalent depth position of the defect in the wall thickness direction.

5. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The extraction of the attenuation time constant in step S3 specifically includes: assuming the voltage signal corresponding to the acquired original voltage attenuation curve is... Taking the natural logarithm of the original voltage decay curve, the logarithmized signal is expressed as follows: ; The plateauing region in the latter part of the logarithmically transformed curve is selected as the fitting interval. The least squares method is used to perform linear fitting on the data within this interval, and the fitting function is of the form: ,in The time variable is calculated from the moment the self-excitation pulse is turned off. This is the initial voltage amplitude constant term obtained from the fitting; By solving the slope of the linear fit The decay time constant of the induced eddy current was calculated. Principal component analysis was used to perform feature dimensionality reduction on the original attenuation curve. Attenuation curves of standard test blocks under different coating thicknesses and different lift-off distances were collected in advance to construct a sample matrix. The covariance matrix and its eigenvectors were calculated. The first eigenvector corresponding to the largest eigenvalue was selected as the principal component projection direction. The attenuation curve vector collected in real time was projected onto this principal component direction to obtain the first principal component component.

6. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The real-time calculation of the equivalent thickness of the stainless steel cladding in step S4 specifically includes: pre-selecting stepped test blocks with the same material grade and cladding thickness specifications as the pipe to be tested, performing pulsed eddy current detection in each known thickness region, and recording the corresponding attenuation time constant. A multinomial regression method was used to establish the mapping relationship between the equivalent thickness of the coating and the decay time constant. The mapping model is as follows: ,in For the equivalent thickness of the coating, These are polynomial weighting coefficients. The order of the polynomial; During online detection, the decay time constant will be extracted in real time. Substituting into the mapping model, the equivalent thickness of the overlay at the current spatial location is calculated. .

7. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The fusion judgment model in step S5 adopts a multi-classifier architecture based on support vector machine. The feature vector input to the model has a total of 12 dimensions, specifically including: the amplitude value of the reflected wave from the heterogeneous interface after gain correction in step S4. Pulse eddy current decay time constant The root mean square value of the residuals of the pulsed eddy current attenuation curve fitting, and the first principal component extracted by principal component analysis; the full width at half maximum (FWHM) of the acoustic echo envelope, the centroid displacement of the acoustic spectrum, and the energy centroid depth of the acoustic signal. The root mean square energy value of the ultrasound A-scan signal in the region below the heterogeneous interface; the peak phase polarity characteristics of the ultrasound A-scan signal in the region below the interface; the spatial gradient of the amplitude of the interface reflection wave between adjacent scan positions; and the attenuation time constant between adjacent scan positions. The spatial gradient of the current scan position; and the location code of the weld area to which the current scan position belongs.

8. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 7, characterized in that, The intelligent identification and positioning in step S5 further includes: the offline training process of the fusion judgment model: collecting samples from the production line including surface cracks on the cladding side, deep cracks on the base layer side, weld non-fusion, weld slag inclusions, local thinning of the cladding, and defect-free normal weld areas, and having professionals perform truth value labeling through destructive testing or radiographic re-examination. Radial basis functions are selected. As the kernel function of the support vector machine, where and They represent the first The and the first The feature vectors of each training sample The kernel function width parameter is used; multiple binary sub-classifiers are constructed using a one-to-many strategy, and the kernel function width parameter is determined through cross-validation. With penalty coefficient Online reasoning and defect classification: For each frame's normalized 12-dimensional feature vector The result is simultaneously fed into multiple trained binary sub-classifiers, with each sub-classifier outputting a decision function value. ,in For the set of support vectors, For the Lagrange multipliers corresponding to the support vectors, For the class labels of support vectors, For support vector features, For bias terms; The class corresponding to the subclassifier with the largest and positive decision function value is selected as the determination result for the current scanning position; where, when the decay time constant... When a shift occurs and the amplitude of the ultrasonic interface reflection wave increases synchronously, it is determined to be a local thinning of the coating. When the pulsed eddy current characteristics are stable but the ultrasonic signal shows high-energy diffraction waves below the heterogeneous interface, it is determined to be a deep crack or lack of fusion on the carbon steel base side; when the full width at half maximum (FWHM) of the ultrasonic echo envelope increases, the spectral centroid shifts to lower frequencies, and the depth of the energy centroid is within the range of the coating thickness, it is determined to be a surface crack or inclusion on the coating side.

9. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The detection frame adopts an adaptive centering structure, with multiple sets of elastic guide wheels distributed at angular intervals straddling both sides of the weld, so that the lift-off distance between the phased array ultrasonic probe and the pulsed eddy current sensor relative to the weld surface is maintained within a preset range. The detection frame is also equipped with an automatic spray coupling system that dynamically adjusts the acoustic coupling agent flow rate according to the detection speed. The data processing unit integrates a general-purpose processor and an FPGA hardware acceleration module. The FPGA hardware acceleration module obtains the gain compensation value by looking up the equivalent thickness deviation calculated in step S4, and writes the control word into the digital variable gain amplifier register inside the phased array controller through the serial peripheral interface to perform real-time scaling of the echo signal amplitude.

10. The intelligent detection method for weld defects in stainless steel composite straight seam welded pipes according to claim 1, characterized in that, The fusion judgment model has online learning capabilities. The system automatically records the defect data of each judgment and its corresponding multimodal raw signal. After the inspection personnel complete the manual review and enter the results, the online learning module compares the differences between the manual judgment and the system judgment, and uses an incremental learning strategy to automatically update the support vectors and corresponding weight coefficients of the support vector machine classifier. The method includes an automatic response mechanism for abnormal situations: during the coordinate establishment phase, if the displacement encoder loses pulses due to slippage, the system senses the motion status through the built-in accelerometer and triggers an early warning in real time, pausing data recording until the coordinates are recalibrated. During data acquisition, if insufficient water supply from the sprinkler system causes acoustic coupling failure, the processing center will identify the overall drop in background wave amplitude and automatically mark it as an invalid detection area in the record. Once coupling is restored, a retest will be prompted.

Images