Ultrasonic nondestructive testing method and system for welds of building steel structures

Abstract

The application discloses a building steel structure weld ultrasonic nondestructive testing and flaw detection method and system, belongs to the technical field of weld detection, and the method comprises the following steps: acquiring weld surface laser point cloud data, calculating local normal direction and converting probe inclination; extracting noise amplitude sequence in a non-welding area, and constructing a noise statistics table; driving an ultrasonic array probe to generate a full matrix echo data set and a plane wave echo data set according to a scanning posture table; mapping into a full focus image and a plane wave superposition image, calculating amplitude difference to generate a fusion weight map; generating a fusion image based on the fusion weight map, performing regional segmentation by using the noise statistics table, and extracting a defect boundary coordinate table. The technical scheme of laser point cloud guided probe posture, dual-mode imaging fusion and adaptive noise segmentation can realize high-precision, automatic identification and objective quantitative evaluation of internal defects of a weld.

Description

Technical Field

[0001] This invention belongs to the field of weld inspection technology, and particularly relates to a method and system for ultrasonic non-destructive testing of welds in building steel structures. Background Technology

[0002] Steel structures are widely used in modern construction projects due to their excellent strength and durability. As the core load-bearing structure of modern construction projects, the quality of the welds in steel structures directly determines the structural safety and durability. Ultrasonic non-destructive testing is the mainstream technology for weld defect detection. Among them, ultrasonic phased array testing is widely used to identify welding defects such as cracks, slag inclusions, and porosity due to its advantages of intuitive imaging and high detection efficiency.

[0003] Current conventional ultrasonic phased array testing methods mainly rely on manually setting detection parameters and completing the scan using a fixed beam angle. Some improved schemes incorporate workpiece surface morphology data for beam parameter compensation, enabling basic defect detection and preliminary assessment. However, in practical engineering applications, steel structure weld surfaces often suffer from issues such as anti-corrosion coatings, rust, oxide scale, and uneven roughness, leading to poor ultrasonic probe coupling, beam energy loss, and unstable echo signals. Furthermore, existing methods often employ a single imaging mode, making it difficult to balance high resolution and high signal-to-noise ratio, limiting the detection of detailed and deep defects. Threshold segmentation also relies on manual experience or fixed reference values, resulting in strong subjectivity and susceptibility to material noise interference, ultimately leading to missed detection of minute defects and misjudgment of complex defects, failing to achieve high-precision, objective quantitative defect assessment. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a method and system for ultrasonic non-destructive testing of weld seams in building steel structures. The system employs a technical solution that uses laser point cloud to guide probe posture, dual-mode imaging fusion, and adaptive noise segmentation, enabling high-precision, automated identification and objective quantitative assessment of internal weld defects.

[0005] The above objectives can be achieved through the following approach:

[0006] The ultrasonic non-destructive testing method for welds in steel structures includes the following steps:

[0007] Acquire laser point cloud data of weld surface, select local neighborhood points to perform plane fitting, calculate local normal direction and convert probe tilt angle, and generate scanning attitude table;

[0008] The ultrasonic array probe is moved to the non-welding area outside the weld to be tested to collect the echo waveform. The structural scattering signal segment between the initial wave and the bottom wave is extracted to extract the noise amplitude sequence. The amplitude histogram and cumulative ratio are calculated using the noise amplitude sequence to generate a noise statistics table.

[0009] The ultrasonic array probe is driven to perform point-by-point positioning according to the scanning attitude table, and the array elements are transmitted and received in sequence through all channels. The data is stored according to the point number to generate a full matrix echo dataset.

[0010] The ultrasonic array probe is driven to perform point-by-point positioning according to the scanning attitude table, perform multi-angle plane wave transmission and full-channel reception, and store the data according to the point number to generate a plane wave echo dataset.

[0011] The full matrix echo dataset is mapped to generate a full-focus image, and the plane wave echo dataset is mapped to generate a plane wave superimposed image. The amplitude difference is calculated between the full-focus image and the plane wave superimposed image to generate a fusion weight map.

[0012] A fused image is generated by weighted summation of a fully focused image and a plane wave superimposed image based on a fusion weight map. A noise statistics table is then used to perform region segmentation on the fused image, extract the boundaries of connected regions, and generate a defect boundary coordinate table.

[0013] Preferably, generating the scanning posture table includes:

[0014] Statistical outlier filtering is performed on the laser point cloud data of the weld surface to remove noise points, and uniform grid downsampling is performed to construct a sparse point cloud coordinate set;

[0015] Traverse the sparse point cloud coordinate set, construct candidate neighborhood points of different orders of magnitude, perform least squares plane fitting and select the minimum fitting residual, and extract the normal vector as the local normal direction.

[0016] Calculate the geometric angle between the local normal direction and the vertical axis of the point cloud coordinate system, map the geometric angle into the probe tilt angle, perform association storage between the sparse point cloud coordinate set and the probe tilt angle, and generate a scanning attitude table.

[0017] Preferably, generating a noise statistics table includes:

[0018] The ultrasonic array probe is moved to the non-welding area outside the weld seam to be tested, ultrasonic pulses are excited and echo waveforms are acquired.

[0019] By traversing the time axis data through the echo waveform, the initial peak value and the bottom peak value of the echo waveform are identified. The waveform data is extracted using the time coordinate index of the initial peak value and the bottom peak value, and a noise amplitude sequence is generated.

[0020] The noise amplitude sequence is divided into discrete amplitude intervals. The number of sampling points falling into each interval is counted to generate an amplitude histogram. The amplitude histogram is then accumulated step by step to generate a cumulative ratio, and a noise statistics table is constructed.

[0021] Preferably, generating the noise amplitude sequence includes:

[0022] The time axis data of the echo waveform is divided into a near-field search interval and a far-field search interval. The local maximum value in the near-field search interval is locked as the initial wave peak value, and the global maximum value in the far-field search interval is locked as the bottom wave peak value.

[0023] The time coordinate indexes of the initial peak value and the bottom peak value are extracted as the starting point and ending point of the truncation, respectively. The intermediate waveform data is extracted by the starting point and the ending point of the truncation, and the absolute value operation is performed to generate a noise amplitude sequence.

[0024] Preferably, the generation of the full matrix echo dataset includes:

[0025] The spatial coordinates and probe tilt angle in the scanning attitude table are analyzed sequentially to generate multi-axis motion control commands, which drive the ultrasonic array probe to move and fit to the current target scanning point.

[0026] The ultrasonic array probe is activated to perform the single-element sequential excitation action, and all array elements are simultaneously controlled to perform synchronous acquisition action to construct a single-frame full matrix data packet;

[0027] Extract the point number from the scan attitude table, establish an index mapping relationship between the point number and the single-frame full matrix data packet, perform serialization and append storage, and generate a full matrix echo dataset.

[0028] Preferably, the generated plane wave echo dataset includes:

[0029] Read the coordinate parameters in the scanning attitude table, drive the ultrasonic array probe to move and position it to the current scanning point, and set the plane wave emission angle;

[0030] The plane wave emission angles are traversed to calculate the emission delay time of the array elements. The ultrasonic array probes are controlled to perform simultaneous excitation according to the emission delay time, and all array elements are controlled to perform full-channel reception to acquire echo data.

[0031] The location number is read using the scanning attitude table, the echo data is associated with the location number and written to the storage medium to generate a plane wave echo dataset.

[0032] Preferably, the calculation of amplitude difference to generate a fusion weight map for the superimposed image of the full-focus image and the plane wave image includes:

[0033] The detection area is discretized, and the amplitude data is mapped and filled to discrete nodes based on the full matrix echo dataset and the plane wave echo dataset to generate a superimposed image of the full focus image and the plane wave image.

[0034] A matrix subtraction operation is performed using a fully focused image and a plane wave superimposed image to obtain an amplitude residual matrix. The amplitude residual matrix is ​​then subjected to an inversion mapping to generate a fused weight map.

[0035] Preferably, mapping amplitude data to discrete nodes includes:

[0036] The two-dimensional coordinate plane is divided based on the detection area. The two-dimensional coordinate plane is meshed according to the spatial resolution step size, and the mesh intersection points are defined as discrete nodes.

[0037] The sound wave propagation path time is calculated based on discrete nodes. Based on the sound wave propagation path time, waveform amplitude is extracted from the full matrix echo dataset and the plane wave echo dataset. Delay superposition operation is then performed to fill the discrete nodes.

[0038] Preferably, generating the defect boundary coordinate table includes:

[0039] The fused image is generated by performing pixel-level multiplication and numerical superposition on the full-focus image and the plane wave superimposed image respectively using the fusion weight map.

[0040] The segmentation benchmark value is found in the noise statistics table according to the preset noise confidence ratio. Threshold segmentation is performed on the fused image to extract the set of pixels whose amplitude exceeds the segmentation benchmark value, forming defective connected regions.

[0041] The outermost contour pixels are identified based on the connected regions of the defect. The spatial coordinates of the outermost contour pixels are extracted and arranged in order to generate a defect boundary coordinate table.

[0042] An ultrasonic non-destructive testing system for weld seams in building steel structures, used to implement the above methods, includes:

[0043] The point cloud attitude generation module is used to acquire laser point cloud data of the weld surface, select local neighborhood points to perform plane fitting, calculate the local normal direction and convert the probe tilt angle, and generate a scanning attitude table.

[0044] The noise statistics module is used to control the ultrasonic array probe to move to the non-welding area outside the weld to be tested to collect echo waveforms, extract the structural scattering signal segment between the initial wave and the bottom wave to extract the noise amplitude sequence, use the noise amplitude sequence to calculate the amplitude histogram and cumulative ratio, and generate a noise statistics table.

[0045] The full matrix echo acquisition module is used to drive the ultrasonic array probe to perform point-by-point positioning according to the scanning attitude table, perform array element successive transmission and full-channel reception, and store the data according to the point number to generate a full matrix echo dataset.

[0046] The plane wave echo acquisition module is used to drive the ultrasonic array probe to perform point-by-point positioning according to the scanning attitude table, perform multi-angle plane wave transmission and full-channel reception, and store the data according to the point number to generate a plane wave echo dataset.

[0047] The dual imaging weight generation module is used to map the full matrix echo dataset to generate a full-focus image, and to map the plane wave echo dataset to generate a plane wave superimposed image. It also calculates the amplitude difference between the full-focus image and the plane wave superimposed image to generate a fusion weight map.

[0048] The fusion segmentation boundary output module is used to perform weighted summation on the full-focus image and the plane wave superimposed image based on the fusion weight map to generate a fused image, perform region segmentation on the fused image using a noise statistics table, extract the boundaries of connected regions, and generate a defect boundary coordinate table.

[0049] The present invention has the following advantages:

[0050] This invention accurately acquires the actual three-dimensional shape of the weld by laser point cloud scanning, and generates a scanning attitude table accordingly to guide the attitude of the ultrasonic array probe in real time. This ensures that the ultrasonic beam is always incident on the irregular weld surface at the optimal angle, improves the stability of acoustic coupling and the quality of original signal acquisition, and thus guarantees the accuracy and high repeatability of the detection results from the source.

[0051] This invention combines full-matrix acquisition and plane wave acquisition modes, generating high-resolution fully focused images and high signal-to-noise ratio plane wave superimposed images respectively. By calculating the amplitude difference between the two images to create an adaptive fusion weight map, complementary image fusion is achieved, improving the final imaging quality, displaying details of minute defects and accurately outlining the contours of complex defects, thus enhancing the comprehensive detection and characterization capabilities of defects.

[0052] This invention establishes an adaptive segmentation mechanism based on the background noise of actual workpieces. By collecting signals in non-welding areas and constructing a noise statistics table, a segmentation benchmark closely related to the material properties of the workpiece under test is obtained, replacing the subjective threshold set by traditional experience or fixed reference blocks. This makes the defect identification process more objective and reliable, and effectively reduces missed detections and misjudgments caused by material differences. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the method of the present invention;

[0054] Figure 2 This is a statistical distribution characteristic diagram of the noise amplitude of the echo signal from the non-welded area under test in Example 1;

[0055] Figure 3 This is the amplitude correlation distribution of each pixel in the detection area in Example 1 in the full-focus image TFM and the plane wave superimposed image PWI;

[0056] Figure 4 This is a schematic diagram of the system of the present invention. Detailed Implementation

[0057] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings.

[0058] Example 1: As Figure 1 As shown, the ultrasonic non-destructive testing method for welds in building steel structures includes the following steps:

[0059] Acquire laser point cloud data of weld surface, select local neighborhood points to perform plane fitting, calculate local normal direction and convert probe tilt angle, and generate scanning attitude table;

[0060] The ultrasonic array probe is moved to the non-welding area outside the weld to be tested to collect the echo waveform. The structural scattering signal segment between the initial wave and the bottom wave is extracted to extract the noise amplitude sequence. The amplitude histogram and cumulative ratio are calculated using the noise amplitude sequence to generate a noise statistics table.

[0061] The ultrasonic array probe is driven to perform point-by-point positioning according to the scanning attitude table, and the array elements are transmitted and received in sequence through all channels. The data is stored according to the point number to generate a full matrix echo dataset.

[0062] The ultrasonic array probe is driven to perform point-by-point positioning according to the scanning attitude table, perform multi-angle plane wave transmission and full-channel reception, and store the data according to the point number to generate a plane wave echo dataset.

[0063] The full matrix echo dataset is mapped to generate a full-focus image, and the plane wave echo dataset is mapped to generate a plane wave superimposed image. The amplitude difference is calculated between the full-focus image and the plane wave superimposed image to generate a fusion weight map.

[0064] A fused image is generated by weighted summation of a fully focused image and a plane wave superimposed image based on a fusion weight map. A noise statistics table is then used to perform region segmentation on the fused image, extract the boundaries of connected regions, and generate a defect boundary coordinate table.

[0065] The generation of the scan pose table includes:

[0066] Statistical outlier filtering is performed on the laser point cloud data of the weld surface to remove noise points, and uniform grid downsampling is performed to construct a sparse point cloud coordinate set;

[0067] After acquiring high-density point cloud data, statistical outlier filtering is first performed. The principle is that measurement noise is spatially random and sparse, while effective signals exhibit local clustering. The average distance from each point to its U nearest neighbors is calculated, along with the global mean and standard deviation of these average distances. A point is retained only if its average distance is less than the sum of the global mean and standard deviation. This effectively filters out isolated noise points caused by laser diffuse reflection or airborne dust. Subsequently, to improve subsequent computational efficiency, uniform grid downsampling is performed. By establishing a voxel grid with side length L in three-dimensional space, all point cloud data falling within the same voxel grid are replaced with the geometric centroids of all points within that voxel. This preserves the macroscopic geometric features of the weld surface while significantly reducing the number of data points, forming a sparse point cloud coordinate set.

[0068] For example, assume the original point cloud data acquired by the laser contour scanner contains 1 million points. First, the number of nearest neighbors is set to U=50, and the global average distance is calculated to be 0.05mm with a standard deviation of 0.01mm. The standard deviation factor is set to 1.0. At this point, any point with an average neighborhood distance exceeding 0.05 + 1 × 0.01 = 0.06mm is identified as outlier noise and removed. Next, the voxel side length for mesh downsampling is set to L=1mm. After processing, the originally high-density point cloud is reduced to a sparse point cloud coordinate set of approximately 10,000 points. These points are evenly distributed on the weld surface, and burr noise has been removed, providing a data foundation for normal calculation.

[0069] Traverse the sparse point cloud coordinate set, construct candidate neighborhood points of different orders of magnitude, perform least squares plane fitting and select the minimum fitting residual, and extract the normal vector as the local normal direction.

[0070] The core concept lies in "adaptive fitting," which automatically finds the most suitable fitting range for the curvature variations in different regions of the weld surface. For each target point in the sparse point cloud coordinate set, multiple sets of candidate neighborhood points with different orders of magnitude are established. For each set of candidate neighborhood points, the least squares method is used to solve for an optimal fitting plane that minimizes the sum of the squares of the perpendicular distances from all points within that neighborhood to the plane. After calculation, multiple fitting residual values ​​corresponding to different numbers of neighborhoods are obtained. The fitting residual reflects the degree of deviation between the points within the currently selected neighborhood range and the ideal plane: the smaller the residual, the more coplanar the points within that range, and the more the fitted normal represents the true surface orientation of that point. Therefore, the set with the smallest fitting residual is selected, and its corresponding plane normal vector is extracted and defined as the local normal direction of that point.

[0071] For example, consider a transition fillet region of a weld where curvature varies significantly. Three sets of candidate neighborhood points are constructed for the current target point: the first set selects the 10 nearest points, the second set selects 20 points, and the third set selects 50 points. When fitting with 10 points, the range is too small, making it highly susceptible to single-point errors, resulting in a residual of 0.02. When fitting with 20 points, the range is moderate, covering both features and smoothing out errors, resulting in a residual of 0.005. When fitting with 50 points, the range is too large, introducing non-planar curved sections, causing the residual to surge to 0.15. The fitting result corresponding to the second set with the smallest residual is automatically selected, and its normal vector is extracted as the local normal direction for that point. This mechanism avoids the problems of using too few points in flat areas leading to instability, or using too many points in curved areas leading to "smoothing out" features.

[0072] Calculate the geometric angle between the local normal direction and the vertical axis of the point cloud coordinate system, map the geometric angle into the probe tilt angle, perform association storage between the sparse point cloud coordinate set and the probe tilt angle, and generate a scanning attitude table.

[0073] After obtaining the local normal direction, it needs to be converted into a control angle that the robotic arm or scanning frame can execute. This requires calculating the angle between the normal vector and the vertical axis of the global coordinate system, using the following formula:

[0074] ;

[0075] in, It represents the geometric angle, measured in degrees or radians. Physically, it represents the degree of deviation of the weld surface normal from the vertical direction, determining how much the probe needs to be deflected to ensure that the sound beam is incident perpendicularly. This represents the extracted local normal direction vector. , is the output of the least squares plane fitting. The vertical axis vector of the point cloud coordinate system is defined as the Z-axis direction of the global coordinate system. Its value is based on the installation reference of the scanning mechanism, assuming the probe's initial state is vertically downward. After calculating the geometric angle, it is converted into the final probe tilt angle through a linear mapping relationship according to the installation method of the ultrasonic array probe. Finally, the spatial coordinates are... The calculated probe tilt angle is combined with a record and stored in a database or file to form a scanning attitude table.

[0076] For example, suppose that at a point on the weld bevel, the local normal direction vector is obtained through fitting calculation. The vector has been normalized with a magnitude of 1, and is the vertical axis vector of the point cloud coordinate system. The magnitude is 1. Substitute into the formula to calculate: vector dot product. Geometric angle This means that the surface at this point is tilted 30 degrees relative to the horizontal plane. If the ultrasonic array probe is mounted vertically, the probe tilt angle should be set to 30 degrees to maintain perpendicular incidence. The processing module writes the data row {ID:1024,X:55.2,Y:120.5,Z:10.1,Angle:30.0} into the scanning attitude table. When the detection device moves to this coordinate, it will automatically tilt the probe by 30 degrees.

[0077] The generated noise statistics table includes:

[0078] The ultrasonic array probe is moved to the non-welding area outside the weld seam to be tested, ultrasonic pulses are excited and echo waveforms are acquired.

[0079] The scanning mechanism, equipped with a position encoder, moves the ultrasonic array probe to the non-welded area outside the weld to be tested. This non-welded area is typically defined as a base material region at a safe distance from the edge of the weld heat-affected zone, such as 20mm to 50mm, ensuring that this area is free of defects such as porosity or slag inclusions generated during welding, and that its microstructure is consistent with the base material of the weld area. Once at the designated position, the ultrasonic array probe is controlled to operate in pulse-echo mode, exciting high-voltage ultrasonic pulses that penetrate the workpiece. An analog-to-digital converter (ADC) is used to acquire the echo waveform in the time domain at a high sampling rate, such as 50MHz to 100MHz.

[0080] By traversing the time axis data through the echo waveform, the initial peak value and the bottom peak value of the echo waveform are identified. The waveform data is extracted using the time coordinate index of the initial peak value and the bottom peak value, and a noise amplitude sequence is generated.

[0081] The acquired echo waveforms contain the initial wave, the bottom wave, and the scattered signal between them. By traversing the time axis data of the echo waveforms and using amplitude threshold search or gradient detection algorithms, the earliest significant high-amplitude signal on the time axis is identified as the initial wave peak, and the second significant high-amplitude signal that appears subsequently is identified as the bottom wave peak. These two peaks represent the time points when the sound beam enters and leaves the workpiece, respectively. To obtain pure structural noise, the time coordinate index corresponding to the initial wave peak is locked. Time coordinate index corresponding to the bottom wave peak and extract the time interval. The signal scattering is processed by removing the influence of the initial wave and the bottom wave themselves, retaining only the structural scattering signal in the middle section. Then, the absolute value of the amplitude of this truncated data is taken to form a non-negative numerical sequence, i.e., the noise amplitude sequence.

[0082] For example, suppose a steel plate with a thickness of 20mm is being tested, and the speed of sound is 5900m / s. The acquired echo waveform data has a length of 2000 sampling points. Traversing the data, a signal with an amplitude of 32000 is found at the 150th sampling point, which is at full scale of the ADC and identified as the initial peak value; a signal with an amplitude of 28000 is found at the 850th sampling point and identified as the bottom peak value. Then, the data between the 180th and 820th points is extracted. Assuming that a certain segment of the original fluctuation data is [5,-8,12,-3,0], the absolute value operation is performed on it to generate a noise amplitude sequence [5,8,12,3,0].

[0083] The noise amplitude sequence is divided into discrete amplitude intervals. The number of sampling points falling into each interval is counted to generate an amplitude histogram. The amplitude histogram is then accumulated step by step to generate a cumulative ratio, and a noise statistics table is constructed.

[0084] After obtaining the noise amplitude sequence, its statistical distribution characteristics need to be quantified. First, based on the dynamic range of the acquisition device, the amplitude range is divided into M equally spaced discrete amplitude intervals. Then, each data point in the noise amplitude sequence is traversed to determine which interval its value falls into, and the counters for the corresponding intervals are accumulated, thereby generating an amplitude histogram reflecting the frequency distribution of the noise amplitude. To determine the confidence threshold for the noise, the cumulative proportion for each interval is calculated based on the amplitude histogram. The formula for calculating the cumulative proportion is as follows:

[0085] ;

[0086] in, represents the cumulative proportion of the i-th discrete amplitude interval, ranging from 0 to 1. Physically, it represents the probability proportion of noise points with amplitudes less than or equal to the upper limit of the i-th interval among all noise samples. i represents the index number of the currently calculated discrete amplitude interval, increasing from 1 to the total number of intervals M. This represents the number of sampling points in the k-th discrete amplitude interval of the amplitude histogram. This represents the total number of sampling points in the noise amplitude sequence, i.e. Finally, the amplitude values ​​for each discrete amplitude interval are compared with the calculated cumulative ratio. A one-to-one mapping relationship is established to construct a noise statistics table. This table allows the corresponding noise amplitude threshold to be retrieved based on a preset confidence level, such as 95%.

[0087] For example, assume that the noise amplitude sequence has a total of The amplitude range was divided into 5 simplified discrete amplitude intervals using 1 sampling point. The resulting amplitude histogram data was statistically analyzed. The intervals are as follows: Interval 1: amplitude 0-10, containing 500 points; Interval 2: amplitude 11-20, containing 300 points; Interval 3: amplitude 21-30, containing 150 points; Interval 4: amplitude 31-40, containing 40 points; Interval 5: amplitude 41-50, containing 10 points. Calculate the cumulative percentage for each interval using the formula. When i=1: When i=2: When i=3: When i=4: ;when hour: .

[0088] The final noise statistics table is as follows: [{Amplitude: 10, Proportion: 0.50}, {Amplitude: 20, Proportion: 0.80}, {Amplitude: 30, Proportion: 0.95},...]. If a 95% confidence level needs to be set for defect segmentation, referring to the table shows that an amplitude of 30 should be selected as the segmentation benchmark. Figure 2 As shown, the bar chart represents the frequency of sampling points within each discrete amplitude interval, while the line chart represents the cumulative probability distribution of noise amplitude. This curve allows you to directly find the amplitude threshold corresponding to a specified confidence level, such as 95%, as the segmentation benchmark.

[0089] The generated noise amplitude sequence includes:

[0090] The time axis data of the echo waveform is divided into a near-field search interval and a far-field search interval. The local maximum value in the near-field search interval is locked as the initial wave peak value, and the global maximum value in the far-field search interval is locked as the bottom wave peak value.

[0091] First, obtain the nominal thickness of the workpiece and the propagation speed of the ultrasonic wave in the material. Based on these two physical parameters, calculate the theoretical path time for the ultrasonic wave to travel one round trip inside the workpiece, and use this as a reference to set the division points of the time axis. The formula for calculating the theoretical bottom wave arrival time is as follows:

[0092] ;

[0093] in, The theoretical arrival time of the bottom wave is indicated by microseconds, which determines the expected time and location of its occurrence. 'd' represents the nominal thickness of the workpiece in millimeters, a value derived from the workpiece's design drawings. 'v' represents the speed of sound in the medium, measured in meters per second or millimeters per microsecond. For carbon steel, the longitudinal wave velocity is typically taken as 5900 m / s. The coefficient 2 indicates the round-trip path the ultrasonic wave takes from the surface to the bottom and back.

[0094] Based on the calculation The timeline data is divided into two independent search intervals: the near-field search interval, set from zero time to... Within a certain percentage range. This range is specifically used to capture the high-amplitude initial wave signal generated when the ultrasonic array probe couples with the workpiece surface. Within this range, a local maximum search algorithm is performed to identify the first local peak that exceeds the noise threshold, for example, 5% of the full-scale amplitude, and whose amplitude is higher than its immediate and adjacent sampling points, and this is marked as the initial wave peak. The far-field search range is set close to... Within a time frame, for example , This is the preset time offset. Within this interval, since bottom reflection is usually the strongest signal in the area, a global maximum value search is performed to determine the sampling point with the highest amplitude within this interval as the bottom wave peak value.

[0095] For example, assume the workpiece to be tested is a steel plate with a thickness of d = 25 mm, and the longitudinal wave velocity of the ultrasonic wave is v = 5900 m / s. First, calculate the theoretical bottom wave time. Based on this calculation, the near-field search interval is set to 0 to 2 microseconds. Within this interval, the algorithm finds a pulse with an amplitude of 80% of full scale at 0.5 microseconds and identifies it as the initial peak value. Simultaneously, the far-field search interval is set to 6 to 12 microseconds. Within this interval, the algorithm iterates through the data and finds the maximum value at 8.5 microseconds, with an amplitude of 65% of full scale, and identifies it as the bottom peak value.

[0096] The time coordinate indexes of the initial peak value and the bottom peak value are extracted as the starting point and ending point of the truncation, respectively. The intermediate waveform data is extracted by the starting point and the ending point of the truncation, and the absolute value operation is performed to generate a noise amplitude sequence.

[0097] After successfully identifying the initial and final peak values, the corresponding array indices (time coordinate indices) of these two peak points in the discretized time-axis data sequence are read. These two index values ​​are directly used as boundary conditions for signal truncation: the initial peak index serves as the starting point, and the final peak index serves as the ending point. Subsequently, the waveform data in the middle segment between the starting and ending points is extracted from the original echo waveform array through array slicing. This data segment eliminates strong reflection signals from the surface and bottom, and mainly contains weak structural scattering signals caused by grain boundaries and microstructure inhomogeneities within the material. Since the original RF waveform contains alternating positive and negative voltage values, direct histogram statistics are not convenient. Therefore, point-by-point absolute value operations are performed on the extracted waveform data. The logical expression of the absolute value operation is as follows:

[0098] ;

[0099] in, This represents the a-th data point in the noise amplitude sequence. This represents the array of raw echo waveform data. This represents the starting point of the intercept, i.e., the time coordinate index of the initial wave peak. 'a' is an incrementing integer starting from 0, ranging up to the difference between the intercept endpoint and the intercept starting point. Convert negative voltage signals to positive amplitude.

[0100] For example, assume a sampling rate of 100MHz. The initial peak value is at 0.5 microseconds, corresponding to time index 50. The bottom peak value is at 8.5 microseconds, corresponding to time index 850. Extract 800 sampling points between indices 50 and 850. Assume the first 5 values ​​of the extracted middle waveform data are [-15, 20, -5, 0, 12], in millivolts. After performing absolute value calculation, the first 5 values ​​of the generated noise amplitude sequence become [15, 20, 5, 0, 12].

[0101] The generation of the full matrix echo dataset includes:

[0102] The spatial coordinates and probe tilt angle in the scanning attitude table are analyzed sequentially to generate multi-axis motion control commands, which drive the ultrasonic array probe to move and fit to the current target scanning point.

[0103] First, the scanning attitude table in the storage medium is read, and the data is parsed row by row. Each row of data contains the three-dimensional spatial coordinates of the target scanning point. And the probe tilt angle used to maintain perpendicular incidence. The Cartesian coordinates and angle parameters are converted into pulse control signals for the servo motors of each joint, i.e., multi-axis motion control commands. By executing these commands, the mechanical actuator carrying the ultrasonic array probe is driven to move to a designated position and simultaneously rotate to a designated angle. To ensure that the ultrasonic energy can be effectively transmitted to the workpiece, after moving into position, the actuator is controlled to make slight movements along the normal direction, so that the working surface of the ultrasonic array probe is in close contact with the surface of the non-welded area outside the weld to be tested.

[0104] For example, assume a six-axis industrial robot is used as the actuator. Data for point number 101 is read from the scan attitude table: coordinates. The probe is tilted at a 25-degree angle, and the rotation angles of the robot's six joint axes are calculated as follows: The controller sends a command to drive the motor to rotate, so that the ultrasonic array probe reaches the point with a repeatability of 0.5 mm and maintains a 0-degree angle with the surface normal, thus completing the physical positioning.

[0105] The ultrasonic array probe is activated to perform the single-element sequential excitation action, and all array elements are simultaneously controlled to perform synchronous acquisition action to construct a single-frame full matrix data packet;

[0106] Once the probe is positioned stably, the full matrix acquisition mode is activated. This mode aims to acquire raw waveform data containing all possible combinations of transmit and receive paths. For an ultrasonic array probe with N physical elements, the acquisition process follows strict timing logic. Excitation phase: The j-th element is controlled as the sole transmitter, exciting an ultrasonic pulse into the workpiece. Reception phase: At the instant the j-th element is excited, all N elements simultaneously switch to receive mode, and the analog-to-digital converter (ADC) is activated to synchronously record the echo signal. Looping process: The above process is repeated N times until each element has completed one transmission action. Ultimately, a complete FMC loop will generate N×N independent A-mode scan waveform data. These N×N time-domain waveform data are organized into a matrix according to the transmitter and receiver element indices, forming a single-frame full matrix data packet. This data packet contains all acoustic scattering information within the material at that physical point.

[0107] For example, suppose a 64-element ultrasound array probe is used, with a sampling depth set to 2048 time sampling points. During the acquisition process, element 1 first transmits, and elements 1 to 64 receive, obtaining 64 A-scan waveforms; then element 2 transmits, and elements 1 to 64 receive, obtaining another 64 A-scan waveforms; ... finally, element 64 transmits, and elements 1 to 64 receive, obtaining the last 64 A-scan waveforms. A total of 64 × 64 = 4096 A-scan waveforms are acquired. If each sampling point occupies 2 bytes, then the size of this single-frame full matrix data packet is approximately MB.

[0108] Extract the point number from the scan attitude table, establish an index mapping relationship between the point number and the single-frame full matrix data packet, perform serialization and append storage, and generate a full matrix echo dataset.

[0109] After data acquisition, to ensure precise correspondence between the massive acoustic data and the physical locations on the workpiece, a unique point number is extracted from the current row of the scan attitude table. Subsequently, a data structure or key-value pair is constructed, using the point number as the index key and the newly generated single-frame full matrix data packet as the data value to establish an index mapping relationship. Due to the enormous data volume, to ensure storage efficiency and read / write speed, the combined data is serialized into a compact binary bitstream. Finally, the serialized data blocks are appended to a unified file on a high-speed storage medium. As the scan progresses, this file grows continuously, eventually forming a full matrix echo dataset containing data from all scan points.

[0110] For example, assume the current scan point is numbered ID: 1001. This number is written to the header area of ​​the data packet, followed immediately by a 16MB waveform data body. The storage format is HDF5 layered data format, and the path is marked " / WaveformData / Point_1001". When the entire weld scan is completed, containing 1000 points, the generated full matrix echo dataset will be a single file of approximately 16GB in size. The raw echo data from any location can be quickly and randomly read using the point number index.

[0111] The generated plane wave echo dataset includes:

[0112] Read the coordinate parameters in the scanning attitude table, drive the ultrasonic array probe to move and position it to the current scanning point, and set the plane wave emission angle;

[0113] The controller first reads the scanning attitude table stored in memory or hard drive and parses the coordinate parameters corresponding to the current row. Then, it sends commands to the multi-axis motion mechanism to drive the ultrasonic array probe to move and precisely position itself at the current target scanning point. Simultaneously, it adjusts the probe attitude according to the angles in the attitude table to conform to the surface. After the probe is stably aligned, a plane wave emission angle sequence is set. This sequence typically covers the range of possible defect orientations in the area to be detected, such as a fan-shaped region from negative to positive angles, to ensure that multi-angle scattering information can be acquired.

[0114] For example, suppose the target point coordinates recorded in the scan attitude table are Millimeters. The controller drives the robotic arm to move the ultrasonic array probe to this coordinate. Subsequently, the detection process parameters are loaded, and the plane wave emission angle sequence is set as follows: starting angle -30°, ending angle +30°, angle step size 2°. This means that the probe will sequentially perform 31 plane wave emission and acquisition actions at different angles at this physical point.

[0115] The plane wave emission angles are traversed to calculate the emission delay time of the array elements. The ultrasonic array probes are controlled to perform simultaneous excitation according to the emission delay time, and all array elements are controlled to perform full-channel reception to acquire echo data.

[0116] To synthesize a plane wavefront with a specific directionality within the workpiece, a precise emission time delay needs to be applied to each element in the array. The required emission delay time for each element is calculated in real-time, iterating through the angle sequence for each plane wave emission angle. The calculation is based on the principle of acoustic path difference, which states that to ensure the emitted wavelets from all elements form a wavefront of the same phase in a specific direction, elements positioned earlier need to emit later, and elements positioned later need to emit earlier. The formula for calculating the emission delay time is as follows:

[0117] ;

[0118] in, This parameter represents the emission delay time of the nth element, typically measured in microseconds. It controls the triggering time of the excitation pulse on the piezoelectric crystal. This represents the physical position coordinates of the center of the nth array element relative to the center of the probe array, in millimeters. These coordinates are obtained by multiplying the element index number by the element spacing. This represents the currently set plane wave emission angle, in degrees, which is the deflection angle of the synthesized wavefront relative to the probe normal. The medium at this point is a steel structure. After calculating the emission delay time of all array elements at the current angle, the controller controls all array elements of the ultrasonic array probe to simultaneously excite according to these time parameters. The waveforms generated by each element are superimposed in the medium, forming a wavefront... Plane waves propagating at an angle. After excitation, the controller immediately switches the circuit to receiving mode, controlling all array elements to perform full-channel reception, synchronously acquiring and digitally recording the returned radio frequency signals, i.e., echo data.

[0119] For example, assume a linear array probe with an element spacing p = 0.6 mm, a detection material of carbon steel, and a sound velocity v = 5900 m / s. Current plane wave emission angle. For the 10th array element, assuming the array center is the origin, its relative position coordinates are... Millimeters. Calculate the delay time. The controller will send an excitation pulse with a delay of 176.5 nanoseconds to the 10th element. After performing similar calculations and excitations on all 64 elements, a set of 64-channel echo data corresponding to a 10° incident angle will be acquired.

[0120] The location number is read using the scanning attitude table, the echo data is associated with the location number and written to the storage medium to generate a plane wave echo dataset.

[0121] After each transmit-receive cycle, to ensure data traceability, the unique point number is retrieved from the scan attitude table. Then, the acquired echo data is associated with that point number and the corresponding transmit angle value. This association is typically achieved by writing the point ID and angle value into the file header or metadata area of ​​the data frame. Finally, the encapsulated data block is written to a high-speed storage medium, such as an SSD. This process is repeated at all transmit angles until all angles at one point are acquired, then the process moves to the next point. Ultimately, after traversing all scanned points, a complete plane wave echo dataset containing the raw signals from all positions and angles is generated.

[0122] Calculating the amplitude difference to generate a fusion weight map for a superimposed image of a fully focused image and a plane wave image includes:

[0123] The detection area is discretized, and the amplitude data is mapped and filled to discrete nodes based on the full matrix echo dataset and the plane wave echo dataset to generate a superimposed image of the full focus image and the plane wave image.

[0124] First, based on the geometry of the weld to be tested and the extent of the heat-affected zone, a two-dimensional rectangular region is defined in computer memory as the detection area. Then, this detection area is divided into a C-row, F-column grid according to the spatial resolution step size. Each intersection point in the grid is defined as a discrete node and assigned fixed physical coordinates. After discretization, two imaging algorithms are executed in parallel: full-focus image generation: based on the TFM principle of the full-focus method, for each discrete node, the total path time of the ultrasonic wave from all transmitting array elements to that node and back to all receiving array elements is calculated.

[0125] Based on this time index, the corresponding amplitude is extracted from the full-matrix echo dataset, summed, and then filled into the node to generate a fully focused image. This image is characterized by full-aperture focusing, extremely high spatial resolution, and the ability to clearly depict minute defects. Plane wave superimposed image generation: Based on the principle of plane wave imaging (PWI), for the same discrete node, the delay of the plane wave arriving at the node and returning to the receiving array element at different transmission angles is calculated. Based on this delay, the signal is extracted from the plane wave echo dataset, coherently superimposed, and then filled into the node to generate a plane wave superimposed image. This image is characterized by high transmission energy, excellent signal-to-noise ratio, and sensitivity to deep defects.

[0126] For example, assume the area to be measured is 40 mm × 20 mm. Setting the grid step size to 0.1 mm generates a 400 × 200 discrete node matrix, totaling 80,000 pixels. For coordinates... At a discrete node within a millimeter, traverse 64×64=4096 waveforms in the full matrix echo dataset, extract the amplitude of each waveform at a specific path time, and sum them to obtain the numerical value. Fill in the corresponding positions in the fully focused image. Simultaneously, iterate through the echo data from 31 angles in the plane wave echo dataset, perform time-delay stacking, and obtain the numerical values. Fill in the corresponding positions in the superimposed plane wave image.

[0127] A matrix subtraction operation is performed using a fully focused image and a plane wave superimposed image to obtain an amplitude residual matrix. The amplitude residual matrix is ​​then subjected to an inversion mapping to generate a fused weight map.

[0128] Because the gain mechanisms of the TFM and PWI algorithms differ, the absolute magnitudes of their original amplitudes may differ. Before performing the subtraction operation, maximum value normalization is first performed on both the fully focused image and the plane wave stacked image, mapping the pixel intensity values ​​of the two images to a unified value. Within the interval, the influence of dimensions is eliminated. Then, pixel-level matrix subtraction is performed to calculate the values ​​for each corresponding discrete node. The absolute difference in amplitude is used to generate the amplitude residual matrix. The formula for calculating the amplitude residual is:

[0129] ;

[0130] in, Represents the magnitude residual matrix of the first... Line 1 The element values ​​in the column represent the consistency differences between the two imaging methods at the same location. The smaller the difference, the more likely that both methods have confirmed the signal, indicating high reliability. The larger the difference, the more likely it is an artifact or noise from one of the methods. This indicates the normalized fully focused image at point... The intensity value at that location. This represents the normalized plane wave superposition image at point... The intensity value at each point. Finally, to construct the weights needed for fusion, the residuals need to be inverted. That is, the smaller the residual, the larger the weight should be. The formula for generating the fused weight map through inversion mapping is as follows:

[0131] ;

[0132] in, Points in the fusion weight graph The weighting coefficient, and its range of values. . This represents the maximum element value in the magnitude residual matrix, used to normalize the residuals. Using this formula, when... When approaching 0, the weight Approaching 1, the highlight features of the area are preserved; when When it is very large, the weight If the value approaches 0, the display of that area will be suppressed.

[0133] For example, scenario A: in coordinates At this point, there is a real, tiny pore. The point was identified in the full-focus image, and its amplitude was normalized. The plane wave superposition image also identified this point, and the normalized amplitude was... Calculate the residuals: Assume the global maximum residual. Calculate the weights: This point received extremely high fusion weights, and the signal was completely preserved. Scenario B: At coordinates At this point, plane wave imaging produces a spurious artifact, but the same area is clean in full-focus imaging. (Full-focus image normalized amplitude) Normalized amplitude of plane wave superimposed image Calculate the residuals: Calculate the weights: The weight of this point is 0, and the artifact will be completely "erased" during fusion.

[0134] Based on the full matrix echo dataset and the plane wave echo dataset, the amplitude data is mapped and filled to discrete nodes, including:

[0135] The two-dimensional coordinate plane is divided based on the detection area. The two-dimensional coordinate plane is meshed according to the spatial resolution step size, and the mesh intersection points are defined as discrete nodes.

[0136] First, based on the physical dimensions of the weld and heat-affected zone to be tested, a two-dimensional Cartesian coordinate plane is constructed in computer memory. The origin of this coordinate plane is defined at the probe incident point or the center surface of the weld. Then, according to the spatial resolution step size, this two-dimensional coordinate plane is meshed. The value of this step size typically depends on the wavelength of the ultrasonic wave; according to the Nyquist sampling theorem, the step size should be less than half the wavelength to avoid spatial aliasing. Each intersection point of the resulting mesh is defined as a discrete node, corresponding to a pixel in image processing. The set of all discrete nodes constitutes the pixel array of the final imaging result.

[0137] For example, assume the physical extent of the detection area is 0 to 40 mm horizontally and 0 to 20 mm deep. With a spatial resolution step size of 0.1 mm, the processor divides the X-axis into 400 equal parts and the Z-axis into 200 equal parts, generating a 400×200 matrix with a total of 80,000 discrete nodes. Each node has unique physical coordinates.

[0138] The sound wave propagation path time is calculated based on discrete nodes. Based on the sound wave propagation path time, waveform amplitude is extracted from the full matrix echo dataset and the plane wave echo dataset. Delay superposition operation is then performed to fill the discrete nodes.

[0139] For each discrete node, calculate the total flight time required for the ultrasonic wave to travel from the source to that node and then reflect back to the receiver; this is the sound wave propagation path time. For the full matrix echo dataset, the formula for calculating this time is as follows:

[0140] ;

[0141] in, This represents the sound wave propagation path time, measured in microseconds. It indicates the amount of time required for a signal to focus at the target pixel for a given transmit-receive combination. This represents the spatial coordinates of the currently calculated discrete node, in millimeters. This indicates the horizontal coordinate of the center of the current transmitting element, in millimeters. This represents the x-coordinate of the center of the current receiving array element, in millimeters. v in this formula is in millimeters per microsecond and needs to be obtained through calibration using a standard test block. The first term in the numerator of the formula... The second term represents the geometric distance from the transmitting element to the discrete node; This represents the geometric distance from the discrete node back to the receiving array element. Based on the calculated... The data is converted into a discrete sampling index, and the waveform amplitude at that moment is extracted from the corresponding A-scan waveform in the full matrix echo dataset. For the plane wave echo dataset, the calculation logic is similar, but the transmission path is based on the distance from the plane wavefront to the node. Finally, the amplitudes extracted from all transmit and receive data at that node are algebraically summed, and the accumulated total intensity value is assigned to that discrete node. This process is executed in parallel on all nodes, and the final pixel matrix is ​​either a fully focused image or a plane wave superimposed image.

[0142] For example, assuming the material's sound velocity For coordinates A discrete node in millimeters. Current calculation: Element 1 transmits, element 10 receives. Calculate transmission distance: mm. Calculate the receiving distance: mm. Calculate the sound wave propagation path time: Assuming a sampling rate of 100MHz, the corresponding sampling index is 568. The amplitude is extracted from the 568th point of the waveform "Tx1-Rx10" in the full matrix echo dataset. This process is repeated for all 64×64 combinations. Assuming the sum of the extracted amplitudes from all combinations is 1500V, the pixel value of this discrete node is filled with 1500.

[0143] The generation of the defect boundary coordinate table includes:

[0144] The fused image is generated by performing pixel-level multiplication and numerical superposition on the full-focus image and the plane wave superimposed image respectively using the fusion weight map.

[0145] First, retrieve the fused weighted image, the fully focused image, and the plane wave superimposed image. Since these three images have been spatially registered based on the same discrete node grid during generation, iterate through the coordinates of each pixel. The weighted summation operation is performed, and the mathematical formula for the fusion operation is as follows:

[0146] ;

[0147] in, This indicates the fused image at the pixel level. The final amplitude intensity at that point. It represents the original amplitude intensity of a fully focused image at the same point. Fully focused images typically have extremely high spatial resolution and can clearly depict point defects. This represents the original amplitude intensity at the same point in a plane wave superimposed image. Plane wave superimposed images, through multi-angle coherent superposition, typically have a better signal-to-noise ratio and the ability to detect linear defects. This represents the normalized weight coefficient provided by the fused weight map at that point, with a value range of [value range missing]. This coefficient reflects the degree of consistency between the two imaging results. When near When the image is fully focused, the fused image retains more of the features of the fully focused image; when... When the value is small, through complementary terms Information from the plane wave image is incorporated to smooth artifacts and enhance the saliency of the true signal. For example... Figure 3 As shown, the set of points near the diagonal in the figure represents the high-consistency defect signals identified by both imaging algorithms, while the discrete points off the diagonal represent random noise or single-mode artifacts that will be suppressed by the fusion algorithm.

[0148] For example, assuming in image coordinates A tiny crack tip was detected at that point. At that point, the amplitude of the fully focused image... Amplitude of a plane wave superimposed image The weights of the fused weight map at that point. Perform calculations Ultimately, the intensity of this pixel in the fused image was determined to be 0.895, effectively preserving the high-intensity defect features.

[0149] The segmentation benchmark value is found in the noise statistics table according to the preset noise confidence ratio. Threshold segmentation is performed on the fused image to extract the set of pixels whose amplitude exceeds the segmentation benchmark value, forming defective connected regions.

[0150] To achieve adaptive segmentation of defect areas and avoid misjudgments caused by manually setting fixed thresholds, a noise statistics table is used. This table stores the correspondence between noise amplitude and cumulative proportion. First, the noise confidence ratio is read, for example, 99.5% or... The image is then horizontally segmented, and the amplitude value corresponding to this proportion is found in the noise statistics table and set as the segmentation benchmark. This means that any signal with an amplitude higher than this benchmark has a 99.5% probability that it is not random noise, but a genuine structural reflection. Subsequently, global binarization thresholding is performed on the fused image, marking all pixels with amplitudes exceeding the segmentation benchmark as "target points" and the rest as "background points," thus extracting the pixel set. To combine discrete pixels into physically meaningful defect objects, a connected component analysis algorithm is used to scan the binarized image. If two "target points" are adjacent in the horizontal, vertical, or diagonal direction, they are considered to belong to the same connected component. After traversal, all interconnected pixels are merged into an independent defect connected component.

[0151] The outermost contour pixels are identified based on the connected regions of the defect. The spatial coordinates of the outermost contour pixels are extracted and arranged in order to generate a defect boundary coordinate table.

[0152] For each identified defect-connected region, an edge detection or contour tracking algorithm is invoked to identify the outermost contour pixels that constitute the geometry of that region. These pixels form the boundary between the defect and the background. Subsequently, based on the spatial resolution stride of the generated image, the row and column indices of these contour pixels in the image matrix are determined. Convert to coordinates in physical space The conversion formula is:

[0153] ;

[0154] ;

[0155] Finally, the processor arranges these transformed physical space coordinates in a clockwise or counterclockwise tracing order to ensure the geometric continuity of the boundary, and writes them into a structured data file to generate a defect boundary coordinate table, which directly quantifies the shape, size and location of the defect.

[0156] Example 2: An ultrasonic non-destructive testing system for weld seams in building steel structures, used to implement the method in Example 1, such as... Figure 4 As shown, it includes:

[0157] The point cloud attitude generation module is used to acquire laser point cloud data of the weld surface, select local neighborhood points to perform plane fitting, calculate the local normal direction and convert the probe tilt angle, and generate a scanning attitude table.

[0158] The noise statistics module is used to control the ultrasonic array probe to move to the non-welding area outside the weld to be tested to collect echo waveforms, extract the structural scattering signal segment between the initial wave and the bottom wave to extract the noise amplitude sequence, use the noise amplitude sequence to calculate the amplitude histogram and cumulative ratio, and generate a noise statistics table.

[0159] The full matrix echo acquisition module is used to drive the ultrasonic array probe to perform point-by-point positioning according to the scanning attitude table, perform array element successive transmission and full-channel reception, and store the data according to the point number to generate a full matrix echo dataset.

[0160] The plane wave echo acquisition module is used to drive the ultrasonic array probe to perform point-by-point positioning according to the scanning attitude table, perform multi-angle plane wave transmission and full-channel reception, and store the data according to the point number to generate a plane wave echo dataset.

[0161] The dual imaging weight generation module is used to map the full matrix echo dataset to generate a full-focus image, and to map the plane wave echo dataset to generate a plane wave superimposed image. It also calculates the amplitude difference between the full-focus image and the plane wave superimposed image to generate a fusion weight map.

[0162] The fusion segmentation boundary output module is used to perform weighted summation on the full-focus image and the plane wave superimposed image based on the fusion weight map to generate a fused image, perform region segmentation on the fused image using a noise statistics table, extract the boundaries of connected regions, and generate a defect boundary coordinate table.

[0163] The implementation details of each module are the same as in Example 1.

[0164] It should be noted that the functional division and information interaction between the various modules described above are logical, but in terms of physical implementation, they can be integrated on the same software platform or deployed in a distributed manner. The connections between modules represent data flow and control flow, aiming to collaboratively achieve the objectives of this invention. The above are merely exemplary embodiments of this invention and should not be construed as limiting the scope of protection of this invention.

Claims

1. A method for ultrasonic non-destructive testing of welds in steel structures, characterized by the following steps: include: Acquire laser point cloud data of the weld surface, select local neighborhood points to perform plane fitting, calculate the local normal direction and convert the probe tilt angle, and generate a scanning attitude table, including: Statistical outlier filtering is performed on the laser point cloud data of the weld surface to remove noise points, and uniform grid downsampling is performed to construct a sparse point cloud coordinate set; Traverse the sparse point cloud coordinate set, construct candidate neighborhood points of different orders of magnitude, perform least squares plane fitting and select the minimum fitting residual, and extract the normal vector as the local normal direction. Calculate the geometric angle between the local normal direction and the vertical axis of the point cloud coordinate system, map the geometric angle into the probe tilt angle, perform association storage between the sparse point cloud coordinate set and the probe tilt angle, and generate a scanning attitude table; The ultrasonic array probe is moved to the non-welded area outside the weld to be tested to collect echo waveforms. A segment of the structural scattering signal between the initial wave and the bottom wave is extracted to obtain the noise amplitude sequence. The amplitude histogram and cumulative ratio are calculated using the noise amplitude sequence to generate a noise statistics table, including: The ultrasonic array probe is moved to the non-welding area outside the weld seam to be tested, ultrasonic pulses are excited and echo waveforms are acquired. By traversing the time axis data through the echo waveform, the initial peak value and the bottom peak value of the echo waveform are identified. The waveform data is extracted using the time coordinate index of the initial peak value and the bottom peak value, and a noise amplitude sequence is generated. The noise amplitude sequence is divided into discrete amplitude intervals. The number of sampling points falling into each interval is counted to generate an amplitude histogram. The amplitude histogram is then accumulated step by step to generate a cumulative ratio, and a noise statistics table is constructed. The generated noise amplitude sequence includes: The time axis data of the echo waveform is divided into a near-field search interval and a far-field search interval. The local maximum value in the near-field search interval is locked as the initial wave peak value, and the global maximum value in the far-field search interval is locked as the bottom wave peak value. The time coordinate indexes of the initial wave peak and the bottom wave peak are extracted as the starting point and ending point of the truncation, respectively. The intermediate waveform data is extracted by the starting point and the ending point of the truncation, and the absolute value operation is performed to generate a noise amplitude sequence. The ultrasonic array probe is driven to perform point-by-point positioning according to the scanning attitude table, and the array elements are transmitted and received in sequence through all channels. The data is stored according to the point number to generate a full matrix echo dataset. The ultrasonic array probe is driven to perform point-by-point positioning according to the scanning attitude table, perform multi-angle plane wave transmission and full-channel reception, and store the data according to the point number to generate a plane wave echo dataset. The full-matrix echo dataset is mapped to generate a fully focused image, and the plane wave echo dataset is mapped to generate a plane wave superimposed image. Amplitude differences are calculated between the fully focused image and the plane wave superimposed image to generate a fusion weight map, including: The detection area is discretized, and the amplitude data is mapped and filled to discrete nodes based on the full matrix echo dataset and the plane wave echo dataset to generate a superimposed image of the full focus image and the plane wave image. A matrix subtraction operation is performed using a fully focused image and a plane wave superimposed image to obtain an amplitude residual matrix. An inversion mapping is then performed on the amplitude residual matrix to generate a fused weight map. Mapping amplitude data to discrete nodes includes: The two-dimensional coordinate plane is divided based on the detection area. The two-dimensional coordinate plane is meshed according to the spatial resolution step size, and the mesh intersection points are defined as discrete nodes. The sound wave propagation path time is calculated based on discrete nodes. Based on the sound wave propagation path time, waveform amplitude is extracted from the full matrix echo dataset and the plane wave echo dataset. Delay superposition operation is performed to fill the discrete nodes. A fused image is generated by weighted summation of the fully focused image and the plane wave superimposed image based on a fusion weight map. Region segmentation is performed on the fused image using a noise statistics table to extract connected region boundaries and generate a defect boundary coordinate table, including: The fused image is generated by performing pixel-level multiplication and numerical superposition on the full-focus image and the plane wave superimposed image respectively using the fusion weight map. The segmentation benchmark value is found in the noise statistics table according to the preset noise confidence ratio. Threshold segmentation is performed on the fused image to extract the set of pixels whose amplitude exceeds the segmentation benchmark value, forming defective connected regions. The outermost contour pixels are identified based on the connected regions of the defect. The spatial coordinates of the outermost contour pixels are extracted and arranged in order to generate a defect boundary coordinate table.

2. The ultrasonic non-destructive testing method for welds in building steel structures according to claim 1, characterized in that, The generation of the full matrix echo dataset includes: The spatial coordinates and probe tilt angle in the scanning attitude table are analyzed sequentially to generate multi-axis motion control commands, which drive the ultrasonic array probe to move and fit to the current target scanning point. The ultrasonic array probe is activated to perform the single-element sequential excitation action, and all array elements are simultaneously controlled to perform synchronous acquisition action to construct a single-frame full matrix data packet; Extract the point number from the scan attitude table, establish an index mapping relationship between the point number and the single-frame full matrix data packet, perform serialization and append storage, and generate a full matrix echo dataset.

3. The ultrasonic non-destructive testing method for welds in building steel structures according to claim 1, characterized in that, The generated plane wave echo dataset includes: Read the coordinate parameters in the scanning attitude table, drive the ultrasonic array probe to move and position it to the current scanning point, and set the plane wave emission angle; The plane wave emission angles are traversed to calculate the emission delay time of the array elements. The ultrasonic array probes are controlled to perform simultaneous excitation according to the emission delay time, and all array elements are controlled to perform full-channel reception to acquire echo data. The location number is read using the scanning attitude table, the echo data is associated with the location number and written to the storage medium to generate a plane wave echo dataset.

4. An ultrasonic non-destructive testing system for weld seams of building steel structures, used to implement the ultrasonic non-destructive testing method for weld seams of building steel structures as described in any one of claims 1-3, characterized in that, include: The point cloud attitude generation module is used to acquire laser point cloud data of the weld surface, select local neighborhood points to perform plane fitting, calculate the local normal direction and convert the probe tilt angle, and generate a scanning attitude table. The noise statistics module is used to control the ultrasonic array probe to move to the non-welding area outside the weld to be tested to collect echo waveforms, extract the structural scattering signal segment between the initial wave and the bottom wave to extract the noise amplitude sequence, use the noise amplitude sequence to calculate the amplitude histogram and cumulative ratio, and generate a noise statistics table. The full matrix echo acquisition module is used to drive the ultrasonic array probe to perform point-by-point positioning according to the scanning attitude table, perform array element successive transmission and full-channel reception, and store the data according to the point number to generate a full matrix echo dataset. The plane wave echo acquisition module is used to drive the ultrasonic array probe to perform point-by-point positioning according to the scanning attitude table, perform multi-angle plane wave transmission and full-channel reception, and store the data according to the point number to generate a plane wave echo dataset. The dual imaging weight generation module is used to map the full matrix echo dataset to generate a full-focus image, and to map the plane wave echo dataset to generate a plane wave superimposed image. It also calculates the amplitude difference between the full-focus image and the plane wave superimposed image to generate a fusion weight map. The fusion segmentation boundary output module is used to perform weighted summation on the full-focus image and the plane wave superimposed image based on the fusion weight map to generate a fused image, perform region segmentation on the fused image using a noise statistics table, extract the boundaries of connected regions, and generate a defect boundary coordinate table.

Images