Ultrasonic three-dimensional nondestructive testing method based on multi-probe echo data joint calibration

Abstract

The invention discloses an ultrasonic three-dimensional nondestructive testing method based on multi-probe echo data joint calibration, and relates to the technical field of material detection, and the ultrasonic three-dimensional nondestructive testing method comprises the following steps: S1, multi-source data space collaborative calibration; s2, echo signal feature extraction and joint optimization; s3, performing multi-view synthetic aperture three-dimensional imaging; and S4, three-dimensional defect automatic identification and quantification. According to the ultrasonic three-dimensional nondestructive testing method based on multi-probe echo data joint calibration, probe poses and acoustic axis errors are corrected by introducing a reference reflection module, and signal consistency is improved by combining empirical mode decomposition, depth gain compensation and cross-correlation phase alignment; and high-precision defect reconstruction and quantification are realized by adopting angle weighted full-focus imaging and three-dimensional watershed segmentation, so that the problem of insufficient three-dimensional imaging precision caused by space reference deficiency, echo inconsistency and unreasonable fusion weight of a multi-probe system can be effectively solved.

Description

Technical Field

[0001] This invention relates to the field of materials testing technology, and in particular to an ultrasonic three-dimensional non-destructive testing method based on joint calibration of multi-probe echo data. Background Technology

[0002] With the increasing demands for structural integrity in advanced manufacturing and major equipment, ultrasonic nondestructive testing technology plays an irreplaceable role in key fields such as aerospace, rail transportation, and energy equipment. Its core objective—the accurate location and quantitative assessment of internal defects in materials—is directly related to the safe service life and life prediction of components. To meet the high-resolution detection requirements of complex geometries and minute defects, the development of high-precision and robust three-dimensional ultrasonic imaging methods has become a key technological direction in the industry.

[0003] Among them, the ultrasonic three-dimensional inspection method based on multi-probe collaboration focuses on fusing multi-view echo data and realizing three-dimensional reconstruction of defect morphology through spatial calibration and signal joint processing. This technical approach aims to overcome the bottleneck of limited field of view and insufficient imaging resolution of single-probe scanning and build a high-fidelity three-dimensional inspection capability for complex working conditions.

[0004] Existing technologies for multi-probe ultrasonic 3D imaging still have several limitations: First, multi-probe systems lack a unified spatial reference, and probe installation errors and acoustic axis deviations can easily introduce geometric mismatches, affecting voxel mapping accuracy. Second, multi-channel echo signals are often affected by noise, attenuation, and phase inconsistencies; without collaborative optimization, this will reduce the imaging signal-to-noise ratio and consistency. Finally, traditional synthetic aperture imaging strategies have relatively simplified weighting for multi-view data fusion, failing to fully consider the influence of the acoustic beam incident angle on reflection intensity, which may introduce artifacts or weaken real defect features. These factors limit the application effectiveness of multi-probe ultrasonic systems in high-precision 3D quantitative detection scenarios.

[0005] Therefore, a three-dimensional ultrasonic nondestructive testing method based on joint calibration of multi-probe echo data is proposed to solve the above problems. Summary of the Invention

[0006] The main objective of this invention is to provide an ultrasonic three-dimensional nondestructive testing method based on joint calibration of multi-probe echo data, so as to solve the problems mentioned in the background above.

[0007] To achieve the above objectives, the technical solution adopted by this invention is as follows: an ultrasonic three-dimensional non-destructive testing method based on joint calibration of multi-probe echo data, comprising the following steps: S1. Multi-source data spatial collaborative calibration: Multiple ultrasonic probes are arranged on the surface of the workpiece to be tested, and a preset reference reflection module is introduced. The spatial pose deviation and acoustic axis pointing error of each probe relative to a unified coordinate system are obtained through this module, and the probe installation position and acoustic beam direction are dynamically compensated. S2. Echo signal feature extraction and joint optimization; S201. Perform empirical mode decomposition on the acquired multi-channel raw echo signal to separate high-frequency noise components and structural clutter components, and retain the intrinsic mode functions containing defect information. S202. Based on the attenuation law of sound waves propagating in materials, a depth-dependent gain compensation function is constructed to correct the echo amplitude of different depth layers point by point. S203. The cross-correlation algorithm is used to calculate the maximum similarity offset between the echo signals of adjacent probes. Based on this offset, the time axis of each channel signal is shifted to align the multi-view echoes in phase. S3, Multi-view synthetic aperture 3D imaging; S301. Divide the area to be detected into a regularly arranged three-dimensional voxel grid, with each voxel unit having unique spatial coordinates; S302. Based on the principle of full-focus imaging, combined with the calibrated probe spatial parameters, calculate the round-trip acoustic path from each probe to each voxel unit, and extract the amplitude response at that moment from the echo of the corresponding channel. S303. Set weighting coefficients according to the angle between the probe acoustic axis and the line connecting the voxel center, and perform angle-weighted superposition of voxel amplitudes from different viewpoints to generate a high-fidelity three-dimensional reconstructed image. S4. Automatic identification and quantification of three-dimensional defects; S401. Perform gradient enhancement processing on the 3D reconstructed image to highlight the boundary difference between the defect area and the background. S402. Apply the three-dimensional watershed segmentation algorithm to divide the enhanced image into regions and extract the connected regions corresponding to all local maxima as candidate defect bodies. S403. Calculate the spatial volume, principal axis length, and centroid coordinates of each candidate defect body, and output a structured defect report.

[0008] Furthermore, the reference reflection module described in S1 is composed of a high acoustic impedance metal block with grooves or stepped structures of known geometric dimensions machined on its surface, and is fixed to the edge region of the workpiece to be tested; each ultrasonic probe is attached to the workpiece surface by a coupling agent and its initial installation position is defined by a mechanical clamp; the probe array adopts a ring or matrix layout, and the distance between adjacent probes is less than half a wavelength to satisfy the spatial sampling theorem; all probes are connected to the same synchronous trigger controller, which outputs excitation pulses with nanosecond-level precision to ensure that the start time of echo acquisition in each channel is consistent.

[0009] Furthermore, the empirical mode decomposition described in S201 adopts an adaptive screening strategy, and the iteration termination condition is that the standard deviation of the residual signal is lower than a preset threshold; after decomposition, the first N eigenmode functions are retained, where N is determined by the signal energy concentration; the gain compensation function described in S202 adopts an exponential form, and its attenuation coefficient is pre-calibrated according to the material type and frequency response characteristics; the cross-correlation algorithm described in S203 is implemented in the frequency domain, and the calculation process is accelerated by fast Fourier transform, with phase alignment accuracy reaching the sub-sampling point level.

[0010] Furthermore, the full-focusing algorithm described in S302 is based on an accurate sound velocity model, and the sound velocity parameters are obtained through bottom echo flight time inversion; the amplitude response of each voxel is obtained by interpolation of several sampling points before and after the theoretical flight time of the corresponding probe; the weighting coefficient described in S303 adopts the form of a cosine function, that is, the weight is equal to the cosine value of the angle between the probe acoustic axis direction vector and the voxel center pointing vector. When the angle approaches 90 degrees, the weight approaches zero, thereby suppressing the specular reflection interference caused by large incident angles.

[0011] Furthermore, S1 also includes a real-time coupling state monitoring step: by analyzing the time interval between the interface wave and the first bottom surface echo, the actual thickness of the coupling layer between the probe wedge and the workpiece surface is inferred; if the thickness deviates from the standard value by more than the tolerance range, the probe channel data is marked as unreliable and is discarded or downweighted in subsequent imaging processes.

[0012] Furthermore, after S201, a secondary denoising process is also included: applying wavelet threshold filtering to the residual signal obtained from empirical mode decomposition, selecting a wavelet basis function that matches the defect scale, suppressing residual random noise through a soft thresholding strategy, while retaining weak but continuous defect echo characteristics.

[0013] Furthermore, the gradient enhancement described in S401 employs anisotropic diffusion filtering to smooth the internal texture while maintaining the sharpness of the defect edges; the three-dimensional watershed algorithm described in S402 introduces morphological marker control to avoid over-segmentation, with the marker points determined jointly by the local curvature extrema and the grayscale peak value; the defect principal axis length described in S403 is calculated using principal component analysis, and the volume is obtained by multiplying the voxel count by the spatial resolution of a single voxel.

[0014] An ultrasonic three-dimensional non-destructive testing system based on joint calibration of multi-probe echo data includes a multi-probe ultrasonic sensor array, a high-precision synchronous excitation and acquisition unit, an embedded signal processing platform, and a three-dimensional visualization terminal; The multi-probe ultrasonic sensing array consists of multiple broadband piezoelectric transducers, each encapsulated in a temperature-compensated wedge and pressed against the workpiece surface by a spring loading mechanism to ensure coupling stability; the probes maintain a fixed geometric relationship through rigid brackets and integrate a miniature pose sensor for initial installation and calibration. The high-precision synchronous excitation and acquisition unit includes a multi-channel high-voltage pulse generator and a high-speed analog-to-digital converter module. All channels share the same crystal clock source. The jitter of the excitation pulse rising edge is less than ten picoseconds. The sampling rate of the analog-to-digital converter is not less than one million times per second, and the bit width is not less than sixteen bits. The embedded signal processing platform is equipped with a multi-core digital signal processor and runs the aforementioned calibration, optimization and imaging algorithms. The data stream adopts a double-buffered ping-pong mechanism, and the data of the next frame is written to the backup buffer during the processing of the current frame. The 3D visualization terminal receives processing results via gigabit Ethernet, supports volume drawing and slice browsing, and can export STL format defect 3D models for subsequent engineering analysis.

[0015] Furthermore, each probe in the multi-probe ultrasonic sensor array is equipped with an independent preamplifier circuit, and the gain is configurable by software; the high-precision synchronous excitation and acquisition unit connects each probe through a coaxial cable, and the cable length is precisely matched to eliminate transmission delay differences; the embedded signal processing platform has a built-in non-volatile memory for storing material sound velocity parameters, coupling calibration curves, and historical detection templates; the three-dimensional visualization terminal integrates a GPU acceleration module, supporting real-time rotation, scaling, and sectioning operations.

[0016] Furthermore, when the embedded signal processing platform executes step S303, it indexes the pre-stored acoustic axis direction vector according to the probe number, calculates the incident angle in combination with the voxel coordinates, and obtains the corresponding cosine weight by looking up the table; in step S402, the submersion process of watershed segmentation is driven by a priority queue, and the priority is determined by the voxel gray level and the neighborhood gradient to ensure that the segmentation boundary extends along the real defect contour.

[0017] The present invention has the following beneficial effects: 1. This invention introduces a reference reflection module with known geometric features to establish a unified coordinate system, accurately acquires the six-degree-of-freedom spatial pose deviation and acoustic axis pointing error of each probe, and achieves dynamic compensation. Simultaneously, it adds real-time monitoring of coupling status, eliminates or downweights data from abnormal coupling channels, and, combined with a probe layout that satisfies the spatial sampling theorem and nanosecond-level synchronous excitation, completely solves the geometric mismatch problem caused by probe installation errors and acoustic axis deviations in traditional technologies. This effectively improves voxel mapping accuracy, and in actual testing, the positioning error can be controlled within 0.5mm, laying a precise spatial geometric foundation for high-precision three-dimensional imaging.

[0018] 2. This invention uses empirical mode decomposition combined with wavelet thresholding for secondary denoising to accurately separate and filter out various types of noise while retaining weak defect echo characteristics. Based on material attenuation laws, an exponential depth gain compensation function is constructed to point-by-point correct the echo amplitude at different depths, compensating for signal attenuation caused by material absorption and diffusion. In the frequency domain, a cross-correlation algorithm combined with fast Fourier transform achieves sub-sampling-level phase alignment of multi-channel echoes. The entire process of signal joint optimization significantly improves the signal-to-noise ratio and multi-view consistency of the echo signal, making the signal characteristics of deep, minute defects clearly discernible and solving the imaging blurring problem caused by signal inconsistency in traditional techniques.

[0019] 3. This invention abandons the simplified fusion weighting strategy of traditional synthetic aperture imaging and adopts an angle-weighted full-focusing imaging method. The cosine of the angle between the probe acoustic axis and the voxel center is used as the weighting coefficient. When the incident angle approaches 90 degrees, the weight approaches zero, effectively suppressing specular reflection interference caused by large incident angles. At the same time, by combining the calibrated probe spatial parameters and the accurate sound velocity model derived from the bottom echo, the round-trip sound path from the probe to the voxel is accurately calculated and the amplitude response is extracted by interpolation. Then, the voxel amplitudes from multiple viewpoints are weighted and superimposed. This strategy fully considers the influence of the sound beam incident angle on the reflection intensity, and solves the problems of artifact introduction and weakening of real defect features caused by unreasonable weighting in traditional technology. The generated three-dimensional reconstructed image has high fidelity and can clearly restore the real shape of the defect.

[0020] 4. This invention accurately extracts candidate defect bodies through anisotropic diffusion filtering gradient enhancement and morphological marker-controlled 3D watershed segmentation, avoiding over-segmentation and missed detection. Then, through methods such as voxel counting and principal component analysis, it automatically calculates key parameters such as the spatial volume, principal axis length, and centroid coordinates of the defect and outputs a structured defect report. This achieves automated, standardized, and quantitative assessment of defects, transforming detection data into engineering-usable information. It facilitates quick access to defect information for inspection personnel and connects to subsequent finite element analysis and maintenance decision-making processes, enabling the engineering application of inspection results and enhancing the practical value of nondestructive testing. The entire identification and quantification process is automated and seamlessly integrated with preceding imaging steps. From 3D reconstruction to defect report generation, no manual intervention is required, significantly shortening the inspection cycle. It is suitable for online industrial inspection scenarios, improving overall inspection efficiency. Attached Figure Description

[0021] Figure 1 This is a schematic diagram of the overall process of an ultrasonic three-dimensional non-destructive testing method based on joint calibration of multi-probe echo data according to the present invention. Figure 2 This is a schematic diagram of the specific steps in S2 of the ultrasonic three-dimensional nondestructive testing method based on joint calibration of multi-probe echo data in this invention. Figure 3This is a schematic diagram of the specific steps in S3 of the ultrasonic three-dimensional nondestructive testing method based on joint calibration of multi-probe echo data in this invention. Figure 4 This is a schematic diagram of the specific steps in step S4 of the ultrasonic three-dimensional nondestructive testing method based on joint calibration of multi-probe echo data according to the present invention. Detailed Implementation

[0022] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0023] Please refer to Figures 1 to 4 The diagram illustrates an ultrasonic three-dimensional non-destructive testing method based on joint calibration using multi-probe echo data. The workpiece under test is a 50mm thick carbon steel plate with a sandblasted surface to improve coupling stability. A reference reflection module, made of a high acoustic impedance metal block, is fixedly mounted on one edge of the workpiece. The module has a 2mm deep and 1mm wide rectangular groove machined on its upper surface, serving as a known geometric feature for spatial calibration. Multiple ultrasonic probes are pressed against the upper surface of the workpiece by a spring-loaded mechanism and maintained by rigid supports. The probe array uses a 4×4 matrix layout, with an 8mm center-to-center distance between adjacent probes, less than half the wavelength of the 5MHz ultrasonic frequency in carbon steel, satisfying the spatial sampling theorem. All ultrasonic probes are encapsulated in temperature-compensated acrylic wedges, with a water-based coupling agent coated between the bottom of the wedge and the workpiece to ensure effective acoustic energy transmission. Each ultrasonic probe integrates a miniature posture sensor, such as a MEMS inertial measurement unit, to record the attitude angle and position offset during initial installation.

[0024] Further explanation of this embodiment is needed: the multi-probe ultrasonic sensor array consists of 16 broadband piezoelectric transducers, each connected to an independent preamplifier circuit with a gain range of 0dB–60dB, which can be remotely configured via an embedded signal processing platform. The multi-probe ultrasonic sensor array is connected to a high-precision synchronous excitation and acquisition unit via a precisely matched coaxial cable, with the cable length error controlled within ±1cm to eliminate nanosecond-level time delays caused by differences in the transmission path. The high-precision synchronous excitation and acquisition unit includes a 16-channel high-voltage pulse generator and a 16-channel high-speed analog-to-digital converter module. All channels share the same temperature-compensated crystal oscillator clock source, and the excitation pulse... The surge edge jitter is less than 10ps, the analog-to-digital converter sampling rate is 20MS / s, and the bit width is 16bit, ensuring that the time synchronization accuracy of the echo signals of each channel is better than 0.05μs. The high-precision synchronous excitation and acquisition unit communicates with the embedded signal processing platform through a gigabit Ethernet interface. The embedded signal processing platform is equipped with a quad-core DSP processor with a main frequency of 1.2GHz, runs a real-time operating system, and has built-in non-volatile memory for storing material sound velocity parameters, coupling calibration curves, and historical detection templates. The embedded signal processing platform transmits the processing results to a 3D visualization terminal through another gigabit Ethernet link. This terminal is equipped with an NVIDIA RTX3060 GPU and supports volume rendering, multi-plane reconstruction, and STL format export.

[0025] Before the test begins, step S1 is performed: multi-source data spatial collaborative calibration; a unified coordinate system is established with the center of the groove of the reference reflection module as the origin, the Z-axis is perpendicular to the surface of the workpiece and pointing upwards, and the X and Y axes are along the length and width of the workpiece, respectively; each ultrasonic probe emits short-pulse ultrasonic waves towards the workpiece and receives the echo signals from the bottom surface of the groove of the reference reflection module; since the geometry of the groove is known, the theoretical echo arrival time can be calculated; the measured echo time is compared with the theoretical value to deduce the positional deviation of each ultrasonic probe relative to the unified coordinate system. , , Simultaneously, by analyzing the amplitude distribution of the reflected echo from the sidewall of the groove and fitting the direction of the main lobe of the acoustic beam, the acoustic axis pointing error is obtained. , The aforementioned six-degree-of-freedom deviation parameters are written into the calibration database of the embedded signal processing platform for subsequent dynamic compensation. Furthermore, real-time monitoring of the coupling state is performed in S1 by measuring the time interval between the interface wave and the first bottom surface echo. The interface wave is the reflected wave from the interface between the probe wedge and the workpiece, combined with the known sound velocity of the wedge. and the speed of sound of the workpiece Calculate the actual thickness of the coupling layer .like If the deviation from the standard value exceeds ±0.03mm, the corresponding channel is marked as unreliable and will be downweighted in subsequent imaging.

[0026] Next, proceed to step S2: echo signal feature extraction and joint optimization; S201. First, perform empirical mode decomposition on the raw echo signal acquired from each channel; the decomposition process adopts an adaptive filtering strategy, calculating the upper and lower envelope mean values ​​in each filtering iteration until the residual signal is obtained. Standard deviation The value is below the preset threshold of 0.01; after decomposition, several intrinsic mode functions are obtained, and the first N IMFs are retained, where N is determined by the energy concentration criterion: the order corresponding to the first time the cumulative energy proportion exceeds 95% is N; S202. Based on the attenuation law of sound waves propagating in materials, construct a depth-dependent gain compensation function. ,in For depth coordinates, This is the attenuation coefficient; for carbon steel materials, Based on calibration experiments at 5MHz, it was determined that... This function corrects the echo amplitude at each depth layer point by point, compensating for signal attenuation caused by material absorption and diffusion. S203. Calculate the maximum similarity offset between echo signals from adjacent probes using a cross-correlation algorithm. Based on this offset, perform time-axis translation on the signals of each channel to align the phase of the multi-view echoes. Specifically, select echo signals from two adjacent channels. and Applying a fast Fourier transform to it yields its frequency domain representation. and Calculate the cross power spectrum The cross-correlation function is then obtained through inverse fast Fourier transform. ;Pick The time offset corresponding to the maximum value ,right Implement timeline translation Phase alignment is achieved; this process is iteratively executed between all adjacent channel pairs, ultimately ensuring that all 16 channel echoes are precisely synchronized on the time axis, with phase alignment accuracy reaching the subsampling point level.

[0027] After signal optimization is completed, proceed to step S3: multi-view synthetic aperture 3D imaging; S301. The area to be inspected is divided into a regularly arranged three-dimensional voxel mesh, with each voxel unit having unique spatial coordinates; specifically: the area to be inspected inside the workpiece is divided into a regularly arranged three-dimensional voxel mesh, with voxel sizes of 1mm × 1mm × 1mm and a coverage depth range of 0mm–50mm; each voxel unit has unique spatial coordinates. ; S302. Based on the principle of full-focusing imaging and combined with the calibrated probe spatial parameters, calculate the round-trip acoustic path from each probe to each voxel unit, and extract the amplitude response at that moment from the corresponding channel echo; specifically: based on the principle of full-focusing imaging, calculate the round-trip acoustic path from each voxel to each ultrasonic probe; sound velocity parameters Obtained through bottom echo time-of-flight inversion: Measuring the time from transmission to reception of the bottom echo. ,but ,in For the workpiece thickness; for the first Each ultrasonic probe, its calibrated spatial position is as follows: Then to voxels The one-way distance is The round trip time is From the first Extracting from channel echo signal The accurate amplitude response is obtained by using cubic spline interpolation at several sampling points near the time point, such as three points before and after. ; Subsequently, in step S303, weighting coefficients are set based on the angle between the probe acoustic axis and the voxel center. Voxel amplitudes from different viewpoints are then weighted and superimposed to generate a high-fidelity 3D reconstructed image. Specifically, in step S303, weighting coefficients are calculated. For: Set up a probe The acoustic axis direction vector is , Obtained from S1 calibration, voxel center points towards probe. The unit vector is Then the cosine of the included angle Weight ;when When approaching 90°, Approaching zero, effectively suppressing specular reflection interference at large incident angles; ultimately, voxels The composite amplitude is Iterate through all voxels to generate a high-fidelity 3D reconstructed image.

[0028] Finally, perform step S4: Automatic identification and quantification of three-dimensional defects; S401. Perform gradient enhancement processing on the 3D reconstructed image to highlight the boundary difference between the defect area and the background. Specifically, firstly, apply anisotropic diffusion filtering to the 3D reconstructed image for gradient enhancement. The diffusion coefficient is dynamically adjusted according to the local gradient magnitude. In the defect edge area, diffusion is suppressed to maintain sharpness, and in the uniform background area, diffusion is enhanced to smooth the texture. S402. Apply the 3D watershed segmentation algorithm to divide the enhanced image into regions and extract all connected regions corresponding to local maxima as candidate defect bodies. Specifically, the steps of the 3D watershed segmentation algorithm are as follows: First, perform morphological opening and closing operations on the enhanced image to eliminate isolated noise. Then, calculate the local curvature using the Hessian matrix and generate a set of marker points by combining the local maxima of grayscale. These marker points serve as the seeds for the watershed algorithm, driving the flooding process. The flooding priority is determined by the voxel grayscale value and the neighborhood gradient. Voxels with higher grayscale and smaller gradients are flooded first, ensuring that the segmentation boundary extends along the true defect contour and avoiding oversegmentation. After segmentation, several connected regions are obtained. Regions with a volume of less than 10 voxels are removed as noise, and the rest are used as candidate defect bodies. S403. Calculate the spatial volume, principal axis length, and centroid coordinates of each candidate defect, and output a structured defect report; specifically, for each candidate defect, count the number of voxels it contains. Multiply by the volume of a single voxel to obtain the spatial volume. Principal component analysis is used to decompose the covariance matrix of all voxel coordinates within the defect. The eigenvector corresponding to the largest eigenvalue is the principal axis direction, and its length is... Principal axis length; centroid coordinates The arithmetic mean of all voxel coordinates; the above parameters are encapsulated into a structured defect report and transmitted to a 3D visualization terminal via the network.

[0029] On the 3D visualization terminal, users can interactively browse the defect area: it supports rotation, scaling, and triorthogonal slicing at any angle, and can export 3D models in STL format for finite element analysis or maintenance decisions; the entire inspection process, from data acquisition to defect report generation, takes about 8 seconds, making it suitable for online industrial inspection scenarios; in this embodiment, the system successfully identified a pore defect with a depth of 25mm and a volume of 12mm³ inside the workpiece to be tested, with a main axis length of 4.2mm and a positioning error of less than 0.5mm, verifying the high-precision 3D reconstruction capability of the present invention under complex conditions such as missing spatial references for multiple probes and poor echo consistency.

[0030] To enable those skilled in the art to fully understand and implement this invention, the specific implementation principles of this invention are further supplemented below with a specific application scenario.

[0031] In practical applications for detecting microcracks inside aero-engine turbine disk forgings, the workpiece to be tested is first placed on a testing platform. The workpiece is a nickel-based high-temperature alloy disk with a diameter of 300 mm and a thickness of 60 mm, and its surface is mechanically polished to ensure coupling consistency. A reference reflection module is fixedly installed on the edge of the workpiece. This module is made of tungsten alloy and has a rectangular groove with a depth of 2 mm and a width of 1 mm machined on it as a geometric reference feature. The multi-probe ultrasonic sensor array consists of 16 5MHz broadband piezoelectric transducers, arranged in a 4×4 matrix on the upper surface of the workpiece. The center-to-center distance between adjacent ultrasonic probes is 7.5 mm, which is less than half of the longitudinal wave wavelength in the nickel-based alloy at this frequency, satisfying the spatial Nyquist sampling condition. Each ultrasonic probe is integrated with a temperature-compensated plexiglass wedge through a spring loading mechanism. The bottom surface of the wedge is coated with a special high-temperature coupling agent to ensure that the sound energy can effectively penetrate the workpiece under operating conditions from room temperature to 150°C.

[0032] The high-precision synchronous excitation and acquisition unit connects each ultrasound probe through a coaxial cable with a length error controlled within ±1cm. All channels share the same temperature-compensated crystal oscillator clock source, which makes the excitation pulse rise edge jitter less than 10ps. The analog-to-digital conversion sampling rate reaches 20MS / s and the bit width is 16bit, thus ensuring that the time synchronization accuracy of the 16-channel echo signal is better than 0.05μs. The embedded signal processing platform is based on a quad-core DSP architecture, which executes subsequent calibration and imaging algorithms in real time and transmits the results to a 3D visualization terminal equipped with an NVIDIA RTX3060 GPU via Gigabit Ethernet.

[0033] Before starting the test, perform step S1: multi-source data spatial collaborative calibration; establish a unified coordinate system with the center of the groove of the reference reflection module as the origin; each ultrasonic probe sequentially emits short-pulse ultrasonic waves and receives the reflected echoes from the bottom of the groove; since the groove depth is known to be 2mm, and considering the sound velocity of nickel-based alloys, approximately 5900m / s, the theoretical echo arrival time can be calculated as follows: ; Measured echo time and Comparison, utilization Back-engineering each ultrasound probe at directional positional deviation Similarly, through the sidewalls of the grooves in and Given the known offset in direction, and combined with the peak position of the echo amplitude, the following calculations can be performed: and Simultaneously, the amplitude distribution of the reflected echo from the groove sidewall at different angles was analyzed, and the direction of the main lobe of the sound beam was fitted to obtain the acoustic axis at the pitch angle. and yaw angle The pointing deviation was measured; the aforementioned six-degree-of-freedom parameters were stored in the calibration database of the embedded signal processing platform for geometric correction in subsequent voxel mapping; furthermore, the time interval between the interface wave and the first bottom echo was measured. Combined with the sound speed of the wedge With the speed of sound of the workpiece Calculate the thickness of the coupling layer ;like If the deviation exceeds the range of 0.1mm ± 0.03mm, a weight attenuation factor of 0.3 is applied to the channel in subsequent imaging to suppress amplitude distortion caused by poor coupling.

[0034] Next, proceed to step S2: echo signal feature extraction and joint optimization; S201. Perform empirical mode decomposition on the raw A-scan signal of each channel, and iterate through adaptive filtering until the residual standard deviation is reached. Several intrinsic mode functions are obtained; according to the energy concentration criterion, the first N-order IMFs with a cumulative energy ratio exceeding 95% for the first time are retained, effectively filtering out high-frequency electronic noise and low-frequency drift components. S202, then construct the depth gain compensation function. The attenuation coefficient The calibration experiment determined the value to be 1.1 Np / mm, corresponding to the measured attenuation at 5 MHz in a nickel-based alloy, for each depth. The echo amplitude is compensated point by point to eliminate signal attenuation caused by material absorption and sound beam diffusion; Next, in step S203, a cross-correlation algorithm is used to calculate the maximum similarity offset between the echo signals of adjacent probes. Based on this offset, the time axis of each channel signal is shifted to align the multi-view echoes in phase. Specifically, this involves shifting the time axis of the echo signals of adjacent channels... and Perform cross-correlation phase alignment: First, perform a fast Fourier transform to the frequency domain, then calculate the cross-power spectrum. Then, the cross-correlation function is obtained by inverse fast Fourier transform. Take the time offset corresponding to its maximum value. ,right Implement subsampling interpolation translation Ultimately, subsampling-level synchronization of the full array echoes on the time axis is achieved, ensuring phase consistency of multi-view signals in synthetic aperture imaging.

[0035] After signal optimization is completed, proceed to step S3: multi-view synthetic aperture 3D imaging; S301. Construct a voxel mesh with a resolution of 1 mm³ in the area to be measured, covering a depth of 0 mm–60 mm. S302, speed of sound Flight time via bottom surface echo Dynamic inversion: ,in The workpiece thickness is 60mm; for each voxel Calculate its up to the th Round-trip acoustic path of an ultrasonic probe ,in These are the actual coordinates of the probe after S1 calibration; from the... Extracting from channel echo The precise amplitude is obtained by interpolating the time near the sampling point using cubic spline interpolation. ; Then, step S303 is performed to calculate the angle weighting coefficients. Set up a probe acoustic axis direction vector The unit vector of the voxel pointing to the probe, obtained from S1 calibration. ,but This weighting mechanism ensures that when the sound beam incident angle... When approaching 90°, Approaching zero, thus effectively suppressing specular reflection artifacts caused by large-angle incident light and improving the contrast of real defects; the final voxel synthesis amplitude is The entire mesh is traversed to generate a 3D reconstructed image.

[0036] Finally, perform step S4: Automatic identification and quantification of three-dimensional defects; S401. Apply anisotropic diffusion filtering to the reconstructed image. The diffusion coefficient is dynamically adjusted with the local gradient magnitude. In the high gradient region, diffusion is suppressed to preserve sharp boundaries, and in the low gradient region, diffusion is enhanced to smooth the background texture. S402. Then, morphological opening and closing operations are performed to remove isolated noise points. Local curvature is then calculated using the Hessian matrix, and seed markers are generated by combining local maxima of grayscale. These markers drive 3D watershed segmentation. The flooding priority is determined by both voxel grayscale and neighborhood gradient: high grayscale and low gradient regions are preferentially marked as the same region to ensure that the segmentation boundary extends along the true defect contour and avoid oversegmentation. After removing connected regions with a volume of less than 10 voxels (i.e., <10 mm³), the remaining candidate defect volumes are parameterized: the number of voxels is multiplied by 1 mm³ to obtain the volume V. S403. Perform principal component analysis on all voxel coordinates. Define the eigenvector corresponding to the largest eigenvalue as the principal axis direction, and its projected length as the principal axis length. The centroid coordinates are the arithmetic mean of all voxel coordinates. The above parameters are encapsulated into a structured report and transmitted to the 3D visualization terminal 10, supporting interactive browsing, triorthogonal slicing, and STL export.

[0037] In this application scenario, the system successfully detected a fatigue crack located at a depth of 32mm and a volume of 8.5mm³, with a main axis length of 3.8mm and a measured positioning error of 0.4mm. This verifies that the present invention effectively overcomes the imaging distortion and defect omission problems caused by traditional methods due to probe installation errors, signal phase mismatch, and uncompensated incident angles by using multi-probe spatial collaborative calibration, echo joint optimization, and angle-weighted synthetic aperture imaging, thus achieving high-fidelity three-dimensional quantitative detection.

[0038] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for ultrasonic three-dimensional nondestructive testing based on joint calibration of multi-probe echo data, characterized in that, Includes the following steps: S1. Multi-source data spatial collaborative calibration: Multiple ultrasonic probes are arranged on the surface of the workpiece to be tested, and a preset reference reflection module is introduced. The spatial pose deviation and acoustic axis pointing error of each probe relative to a unified coordinate system are obtained through this module, and the probe installation position and acoustic beam direction are dynamically compensated. S2. Echo signal feature extraction and joint optimization; S201. Perform empirical mode decomposition on the acquired multi-channel raw echo signal to separate high-frequency noise components and structural clutter components, and retain the intrinsic mode functions containing defect information. S202. Based on the attenuation law of sound waves propagating in materials, a depth-dependent gain compensation function is constructed to correct the echo amplitude of different depth layers point by point. S203. The cross-correlation algorithm is used to calculate the maximum similarity offset between the echo signals of adjacent probes. Based on this offset, the time axis of each channel signal is shifted to align the multi-view echoes in phase. S3, Multi-view synthetic aperture 3D imaging; S4. Automatic identification and quantification of three-dimensional defects.

2. The method according to claim 1, characterized in that, S3 further includes the following steps: S301. Divide the area to be detected into a regularly arranged three-dimensional voxel grid, with each voxel unit having unique spatial coordinates; S302. Based on the principle of full-focus imaging, combined with the calibrated probe spatial parameters, calculate the round-trip acoustic path from each probe to each voxel unit, and extract the amplitude response at that moment from the echo of the corresponding channel. S303. Set weighting coefficients according to the angle between the probe acoustic axis and the line connecting the voxel center, and perform angle-weighted superposition of voxel amplitudes from different viewpoints to generate a high-fidelity three-dimensional reconstructed image.

3. The method according to claim 1, characterized in that, S4 further includes the following steps: S401. Perform gradient enhancement processing on the 3D reconstructed image to highlight the boundary difference between the defect area and the background. S402. Apply the three-dimensional watershed segmentation algorithm to divide the enhanced image into regions and extract the connected regions corresponding to all local maxima as candidate defect bodies. S403. Calculate the spatial volume, principal axis length, and centroid coordinates of each candidate defect body, and output a structured defect report.

4. The method according to claim 1, characterized in that, The reference reflection module described in S1 is composed of a high acoustic impedance metal block with grooves or stepped structures of known geometric dimensions machined on its surface, and is fixed to the edge region of the workpiece to be tested. Each ultrasonic probe is attached to the workpiece surface by a coupling agent and its initial installation position is defined by a mechanical clamp. The probe array adopts a ring or matrix layout, and the distance between adjacent probes is less than half a wavelength to satisfy the spatial sampling theorem. All probes are connected to the same synchronous trigger controller, which outputs excitation pulses with nanosecond-level precision to ensure that the start time of echo acquisition in each channel is consistent.

5. The method according to claim 1, characterized in that, The empirical mode decomposition described in S201 employs an adaptive screening strategy, with the iteration termination condition being that the standard deviation of the residual signal is lower than a preset threshold. After decomposition, the first N eigenmode functions are retained, where N is determined by the signal energy concentration. The gain compensation function described in S202 adopts an exponential form, and its attenuation coefficient is pre-calibrated based on the material type and frequency response characteristics. The cross-correlation algorithm described in S203 is implemented in the frequency domain, accelerating the calculation process through fast Fourier transform, and achieving phase alignment accuracy at the sub-sampling point level.

6. The method according to claim 2, characterized in that, The full-focusing algorithm described in S302 is based on an accurate sound velocity model, and the sound velocity parameters are obtained by inverting the flight time of the bottom echo. The amplitude response of each voxel is obtained by interpolation of several sampling points before and after the theoretical flight time of the corresponding probe. The weighting coefficient described in S303 adopts the form of a cosine function, that is, the weight is equal to the cosine value of the angle between the probe acoustic axis direction vector and the voxel center pointing vector. When the angle approaches 90 degrees, the weight approaches zero, thereby suppressing the specular reflection interference caused by large incident angles.

7. The method according to claim 1, characterized in that, The S1 also includes a real-time coupling state monitoring step: by analyzing the time interval between the interface wave and the first bottom surface echo, the actual thickness of the coupling layer between the probe wedge and the workpiece surface is inferred; if the thickness deviates from the standard value by more than the tolerance range, the probe channel data is marked as unreliable and is discarded or downweighted in subsequent imaging processes.

8. The method according to claim 1, characterized in that, The process after S201 includes a secondary denoising process: applying wavelet threshold filtering to the residual signal obtained from empirical mode decomposition, selecting a wavelet basis function that matches the defect scale, and suppressing residual random noise through a soft thresholding strategy while retaining weak but continuous defect echo characteristics.

9. The method according to claim 3, characterized in that, The gradient enhancement described in S401 employs anisotropic diffusion filtering to smooth the internal texture while maintaining the sharpness of the defect edges; the three-dimensional watershed algorithm described in S402 introduces morphological marker control to avoid oversegmentation, with marker points determined by local curvature extrema and grayscale peak values; the defect principal axis length described in S403 is calculated using principal component analysis, and the volume is obtained by multiplying the voxel count by the spatial resolution of a single voxel.

10. An ultrasonic three-dimensional nondestructive testing system based on joint calibration of multi-probe echo data, applied to the ultrasonic three-dimensional nondestructive testing method based on joint calibration of multi-probe echo data as described in any one of claims 1-9, characterized in that, It includes a multi-probe ultrasonic sensor array, a high-precision synchronous excitation and acquisition unit, an embedded signal processing platform, and a three-dimensional visualization terminal; The multi-probe ultrasonic sensor array consists of multiple broadband piezoelectric transducers, each of which is encapsulated in a temperature-compensated wedge and pressed against the workpiece surface by a spring loading mechanism to ensure coupling stability. The probes are kept in a fixed geometric relationship by rigid brackets, and a miniature pose sensor is integrated for initial installation and verification. The high-precision synchronous excitation and acquisition unit includes a multi-channel high-voltage pulse generator and a high-speed analog-to-digital converter module. All channels share the same crystal clock source. The jitter of the excitation pulse rising edge is less than ten picoseconds. The sampling rate of the analog-to-digital converter is not less than one million times per second, and the bit width is not less than sixteen bits. The embedded signal processing platform is equipped with a multi-core digital signal processor and runs the aforementioned calibration, optimization and imaging algorithms. The data stream adopts a double-buffered ping-pong mechanism, and the data of the next frame is written to the backup buffer during the processing of the current frame. The 3D visualization terminal receives processing results via gigabit Ethernet, supports volume drawing and slice browsing, and can export STL format defect 3D models for subsequent engineering analysis.

Images