Fabricated building quality intelligent detection method and system

Abstract

The invention provides a prefabricated building quality intelligent detection method and system, and relates to the technical field of data processing, and the method comprises the steps: employing a deep residual network to carry out the feature fusion analysis of ultrasonic echo signals, so as to obtain a feature vector corresponding to each signal; according to the feature vectors and a time-frequency domain combined threshold segmentation algorithm, welding defect types are recognized, a defect space coordinate set is marked, a defect coordinate set containing four-dimensional information is formed, and the welding defect types comprise air holes, incomplete fusion and cracks; calculating dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model, wherein the parameters comprise a defect depth gradient, a transverse offset vector and a stress wave attenuation coefficient; and inputting the dynamic offset parameters into a pre-trained welding quality evaluation model to predict the service life of the structure. According to the invention, the residual service life of the fabricated building structure can be evaluated more accurately.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing, and particularly to an intelligent detection method and system for the quality of prefabricated buildings. Background Art

[0002] Prefabricated buildings are assembled from numerous precast components at the construction site. The connection quality between precast components plays a decisive role in the safety and stability of the overall structure. At present, there are some limitations in the traditional quality detection methods for prefabricated buildings.

[0003] For example, taking a construction project of a prefabricated residential community as an example, during the construction process, it was found that there were quality hazards at the welding joints between the ends of some cantilever beams and balcony slabs. Through traditional detection means, it was initially judged that there might be welding defects, but the type, location, and severity of the defects could not be accurately determined. After further detailed detection, it was determined that there was an incomplete fusion defect at this welding joint.

[0004] Incomplete fusion mainly refers to the situation where the filler metal and the base metal are not fused together. The main reasons for this defect may be that the groove is not clean, the welding rod movement speed is too fast, the welding current is too small, the electrode angle is improper, etc. In this case, through investigation and analysis, it was due to improper operation by the worker during the welding process, with the welding rod movement speed being too fast, resulting in incomplete fusion in some areas. This incomplete fusion defect seriously affects the strength and integrity of the welding joint, reduces the load-bearing capacity of the structure, and poses a great risk to the safety of the building structure. If such defects cannot be discovered and properly handled in a timely manner, over time and under the action of various loads on the building structure, this part is very likely to break, thereby triggering serious safety accidents. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide an intelligent detection method and system for the quality of prefabricated buildings, which can more accurately evaluate the remaining service life of prefabricated building structures.

[0006] To solve the above technical problems, the technical solution of the present invention is as follows: In the first aspect, an intelligent detection method for the quality of prefabricated buildings, the method includes: Collecting ultrasonic echo signals of the welding joints between the ends of cantilever beams and balcony slabs; Based on the theoretical installation coordinates of the embedded parts, constructing a reference space grid model, generating a three-dimensional defect distribution map of the welding interface through the time-of-flight inversion algorithm of ultrasonic echo signals, and performing spatial registration with the reference grid model to obtain a defect-reference fusion model; Based on the spatial coordinate mapping relationship of the defect-reference fusion model, a deep residual network is used to perform feature fusion analysis on the ultrasonic echo signal to obtain the feature vector corresponding to each signal; according to the feature vector and the time-frequency domain joint threshold segmentation algorithm, the types of welding defects are identified and the set of defect spatial coordinates is marked to form a defect coordinate set containing four-dimensional information, and the types of welding defects include pores, lack of fusion and cracks. Calculate the dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model, and the parameters include defect depth gradient, lateral offset vector and stress wave attenuation coefficient. Input the dynamic offset parameters into the pre-trained welding quality assessment model to predict the structural life.

[0007] In a second aspect, an intelligent quality inspection system for prefabricated buildings includes: An acquisition module for acquiring ultrasonic echo signals of the welding joint between the end of the cantilever beam and the balcony slab. A fusion module for constructing a reference space grid model based on the theoretical installation coordinates of the embedded parts, generating a three-dimensional defect distribution map of the welding interface through the time-of-flight inversion algorithm of the ultrasonic echo signal, and performing spatial registration with the reference grid model to obtain a defect-reference fusion model. A processing module for performing feature fusion analysis on the ultrasonic echo signal by using a deep residual network based on the spatial coordinate mapping relationship of the defect-reference fusion model to obtain the feature vector corresponding to each signal; according to the feature vector and the time-frequency domain joint threshold segmentation algorithm, identifying the types of welding defects and marking the set of defect spatial coordinates to form a defect coordinate set containing four-dimensional information, and the types of welding defects include pores, lack of fusion and cracks. A calculation module for calculating the dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model, and the parameters include defect depth gradient, lateral offset vector and stress wave attenuation coefficient. A prediction module for inputting the dynamic offset parameters into the pre-trained welding quality assessment model to predict the structural life.

[0008] The above solution of the present invention has at least the following beneficial effects.

[0009] By collecting the ultrasonic echo signals of the welding joint between the end of the cantilever beam and the balcony slab and using the time-of-flight inversion algorithm to generate a three-dimensional defect distribution map of the welding interface, the position of the welding defect in the three-dimensional space can be accurately located. Compared with the traditional detection method, the accuracy of defect detection is greatly improved, and small defects can be effectively identified, avoiding potential safety hazards caused by defect omission. For example, for millimeter-level pore defects that were difficult to detect in the past, this method can clearly show their positions and sizes.

[0010] Using the theoretical installation coordinates of embedded parts to construct a reference spatial grid model, performing spatial registration with the three-dimensional defect distribution atlas to obtain a fusion model, and subsequently conducting feature fusion analysis of the deep residual network based on this model. This series of operations realizes the automation and intelligence of the detection process. Compared with manual detection, it greatly shortens the detection time, improves the detection efficiency, can complete the detection work of a large number of welding joints in a short time, meets the requirements of rapid construction of prefabricated buildings, and can save several times the detection time cost in large prefabricated building projects.

[0011] It can not only identify the types of welding defects, such as pores, lack of fusion, and cracks, but also mark the set of defect spatial coordinates to form a defect coordinate set containing four-dimensional information, and calculate dynamic offset parameters such as defect depth gradient, lateral offset vector, and stress wave attenuation coefficient. These parameters provide rich data support for in-depth analysis of the impact of defects on structural performance, help to comprehensively understand the characteristics of welding defects and their potential hazards to the structure, and thus formulate more targeted repair plans. For example, the development trend of defects in the depth direction can be intuitively judged through the defect depth gradient.

[0012] Inputting the dynamic offset parameters into a pre-trained welding quality assessment model to predict the structural life changes the previous way of judging the structural life only by experience or simple estimation. This method is based on actual detection data and scientific model prediction, and can more accurately evaluate the remaining service life of prefabricated building structures. For example, for a prefabricated building structure with welding defects, predicting its remaining life through this method can provide key references for subsequent maintenance decisions, avoid unnecessary maintenance work too early or too late, and save costs while ensuring building safety. Brief Description of the Drawings

[0013] Figure 1 It is a schematic flow chart of an intelligent detection method for the quality of prefabricated buildings provided by an embodiment of the present invention.

[0014] Figure 2 It is a schematic diagram of an intelligent detection system for the quality of prefabricated buildings provided by an embodiment of the present invention. Detailed Embodiments

[0015] Hereinafter, exemplary embodiments of the present disclosure will be described in more detail with reference to the drawings. Although the exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided so that the present disclosure can be more thoroughly understood and the scope of the present disclosure can be fully conveyed to those skilled in the art.

[0016] As Figure 1As shown in the figure, an embodiment of the present invention provides an intelligent detection method for the quality of prefabricated buildings. The method includes the following steps: Collect the ultrasonic echo signals of the welded joints between the cantilever beam ends and the balcony slabs; Based on the theoretical installation coordinates of the embedded parts, construct a reference space grid model. Generate a three-dimensional defect distribution map of the welding interface through the time-of-flight inversion algorithm of the ultrasonic echo signals, and perform spatial registration with the reference grid model to obtain a defect-reference fusion model; Based on the spatial coordinate mapping relationship of the defect-reference fusion model, use a deep residual network to perform feature fusion analysis on the ultrasonic echo signals to obtain the feature vectors corresponding to each signal; According to the feature vectors and the time-frequency domain joint threshold segmentation algorithm, identify the types of welding defects and mark the defect space coordinate sets to form a defect coordinate set containing four-dimensional information. The welding defect types include pores, lack of fusion, and cracks; Calculate the dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model. The parameters include defect depth gradient, lateral offset vector, and stress wave attenuation coefficient; Input the dynamic offset parameters into a pre-trained welding quality assessment model to predict the structural life.

[0017] In the embodiment of the present invention, through the ultrasonic echo signal acquisition and the time-of-flight inversion algorithm, a three-dimensional defect distribution map of the welding interface is constructed and registered with the reference space grid model, realizing millimeter-level precision positioning of the defect position. Compared with traditional ultrasonic detection, which can only provide two-dimensional cross-sectional information, this method can intuitively present the spatial distribution form of defects inside the welding joints (such as the three-dimensional volume of pores and the extension path of cracks). Combining with the reference model constructed based on the theoretical coordinates of the embedded parts, it can accurately identify the offset error of the defects relative to the design axis, providing a spatial position quantification basis for the defect hazard assessment. For example, for a micro-crack with a length of 2 mm in a certain project, its specific coordinates in the X / Y / Z axis directions and the angle with the theoretical axis can be accurately marked, avoiding potential structural safety hazards caused by misjudgment of the defect position.

[0018] The ultrasonic signal is subjected to feature fusion analysis through a deep residual network (ResNet), combined with a time-frequency domain joint threshold segmentation algorithm, which solves the problems of traditional manual interpretation relying on experience and being prone to missed and misjudged. The model can automatically extract multi-dimensional features of the signal in the time-domain waveform, frequency-domain energy distribution, and time-frequency joint domain (such as stress wave attenuation rate, frequency offset), realizing intelligent classification of three typical defects: pores, lack of fusion, and cracks, and the recognition accuracy is improved by more than 30% compared with traditional methods. Taking the lack of fusion defect as an example, the algorithm can effectively distinguish between local lack of fusion caused by insufficient welding current and interface separation caused by the base metal oxide film by analyzing the reflection energy mutation characteristics and phase shift characteristics of the signal at the defect interface, avoiding misjudging the functional interface as a defect.

[0019] By calculating dynamic parameters such as defect depth gradient, lateral offset vector, and stress wave attenuation coefficient, a mapping relationship between the defect spatial position and the structural mechanical properties is established. Among them, the depth gradient reflects the extension trend of the defect along the thickness direction of the component (such as the rate of crack propagation into the core stress area of the cantilever beam), the lateral offset vector quantifies the degree of deviation of the defect from the theoretical stress axis, and the stress wave attenuation coefficient characterizes the influence of the defect on the structural vibration transmission characteristics. In an actual engineering inspection, through this parameter system, it is found that the stress wave attenuation coefficient of a welding joint of a balcony slab exceeds the threshold. Combining with finite element simulation verification, it is confirmed that the defect causes a 15% decrease in the joint stiffness, providing data support for timely taking reinforcement measures.

[0020] In a preferred embodiment of the present invention, ultrasonic echo signals of the cantilever beam end and the balcony slab welding joint are collected, including: A high-frequency piezoelectric ceramic ultrasonic probe with a center frequency of 5-10 MHz is used to match the acoustic impedance characteristics of the base metal (Q345B steel) of the welding joint. The diameter of the probe wafer is φ6 mm to balance the resolution and penetration ability. For the complex geometric curved surface of the welding joint between the cantilever beam and the balcony slab, a custom-made arc wedge probe with a curvature radius of 80 mm is used to ensure that the coupling area with the detection surface (weld reinforcement ≤ 3 mm) is ≥ 90%, reducing boundary reflection noise. A multi-channel ultrasonic acquisition instrument is configured, supporting 16-channel synchronous acquisition, with a sampling rate of 200 MHz and a dynamic range of 120 dB. It integrates a high-precision positioning module (accuracy ±0.1 mm), and the distance between the probe and the detection surface is calibrated in real time through a laser rangefinder to ensure that the acoustic path measurement error < 0.5%.

[0021] The sandblasting process (abrasive is 80-mesh alumina) is used to remove the oxide scale and spatter on the surface of the welding joint, and the roughness is controlled at Ra ≤ 6.3 μm. The detection area is wiped with anhydrous ethanol to remove oil stains and dried, ensuring that the coupling agent (silicone-based gel, acoustic impedance 1.5×10 6 Rayl) is evenly applied with a thickness of 0.2-0.5 mm to avoid bubble generation.

[0022] Collect the theoretical installation coordinates of the embedded parts using a total station (Leica TS60, angular measurement accuracy of 0.5″). Take the axis of the cantilever beam as the Z-axis to establish a right-handed Cartesian coordinate system. Paste reflective marking points with a diameter of 5 mm on the detection surface as the spatial positioning reference, and construct a detection grid with millimeter-level accuracy (grid spacing 5 mm×5 mm).

[0023] Adopt a "zigzag" linear scanning mode, with a step spacing of 2 mm along the weld length direction (Z-axis), and laterally (X-axis) covering a 20-mm area on both sides of the weld (including the heat-affected zone). For complex areas of fillet welds, add a 45° oblique incidence scan (probe inclination angle α = 30° / 60°) to excite transverse and longitudinal wave modes and cover the areas sensitive to lack of fusion defects. Acquisition parameter settings: Excitation mode: Pulse echo method, emission voltage 150 V, pulse width 0.1 μs; Gain control: Initial gain 60 dB, dynamically adjusted in combination with the real-time A-scan waveform to ensure that the bottom echo amplitude reaches 80% of the full screen; Synchronous trigger: Trigger acquisition through an encoder, and synchronously collect 1 set of signals (including 1024 sampling points) every 0.5 mm of movement; Filter settings: 5 - 15 MHz band-pass filter to suppress low-frequency noise and high-frequency clutter.

[0024] During the acquisition process, the A-scan waveform is displayed in real time, and key monitoring is carried out on: the initial wave width (≤0.5 μs) to judge the coupling stability; the amplitude fluctuation of the defect wave (≤5%) to avoid signal attenuation differences caused by changes in probe pressure; the signal-to-noise ratio of the bottom echo (≥20 dB) to ensure effective penetration of the welded fusion zone; when the signal-to-noise ratio of 3 consecutive scan points < 15 dB, automatically trigger the surface reprocessing process (reapply the coupling agent or adjust the probe angle). System calibration and calibration: Use a CSK-IA standard test block (sound velocity 5900 m / s), and by measuring the echo sound path of 20-mm / 50-mm flat-bottomed holes, correct the instrument sound velocity parameter to 5880 ± 10 m / s (considering the influence of the on-site temperature of 25 ± 2℃).

[0025] Probe delay calibration, using a comparison test block with a thickness of 10 mm, measure the time difference of the bottom echo, and correct the probe front distance to within 0.3 mm.

[0026] Sensitivity calibration, raise the echo of the φ2-mm long transverse hole to 80% of the full screen as the defect detection reference sensitivity. Data acquisition and spatio-temporal correlation recording: Synchronous recording at each detection point: the original time-domain signal (1024 points / channel, including amplitude and time parameters); spatial coordinates (X, Y, Z, accuracy ±0.1 mm); detection timestamp (accurate to the millisecond level); environmental parameters (temperature 25±2 °C, humidity 40±5%RH); establish a one-to-one mapping relationship between each A-scan signal and the spatial coordinates to generate a signal matrix file with geographical coordinates, and at the same time record device parameters such as the probe number and the type of coupling agent to form a traceable original detection data set. Among them, for the convenience of use, before specific applications, it is also necessary to preprocess and standardize the ultrasonic echo signal.

[0027] In a preferred embodiment of the present invention, a reference space grid model is constructed based on the theoretical installation coordinates of the embedded parts, including: Extract the theoretical coordinates of the embedded parts from the architectural design drawings, including the coordinates of each embedded part in the three-dimensional space; according to the theoretical coordinates of the embedded parts, determine the spatial range where the grid needs to be constructed, specifically including: directly read the three-dimensional coordinate information of each embedded part from the architectural design drawings (such as BIM models or CAD drawings), and the coordinates are marked in millimeters, including the specific position (X, Y, Z) of the geometric center of the embedded part in the space. For example, by obtaining the center coordinates of a certain embedded part as (1500, 800, 200) through the drawing markings, it means that the embedded part is 1500 mm away from the origin in the length direction (X-axis) of the cantilever beam, 800 mm in the width direction (Y-axis) of the balcony slab, and 200 mm in the thickness direction (Z-axis) of the welding interface. If there are multiple embedded parts, record the coordinates of each embedded part one by one to form a coordinate set (such as embedded part A: 1200, 600, 180; embedded part B: 1800, 1000, 220, etc.). Determine the grid space range: X-axis range: Traverse the X coordinates of all embedded parts to find the minimum value (such as 1200 mm) and the maximum value (such as 1800 mm). Finally, the X-axis range is determined to be from the minimum value to the maximum value, that is, 1200 mm to 1800 mm, to ensure that the distribution of all embedded parts in the length direction of the cantilever beam is covered.

[0028] Y-axis range: Similarly, extract the minimum value (such as 600 mm) and the maximum value (such as 1000 mm) of the Y coordinates, and determine the Y-axis range to be 600 mm to 1000 mm to cover the distribution of the embedded parts in the width direction of the balcony slab.

[0029] Z-axis range: Considering the designed thickness of the welding interface (such as the theoretical thickness is 100 mm), with the average value of the Z coordinates of the embedded parts (such as 200 mm) as the center, expand half of the thickness (±50 mm) to both sides. Finally, the Z-axis range is 150 mm to 250 mm to ensure that the complete thickness area of the welding interface is included.

[0030] Set the size of the grid cells according to the resolution of ultrasonic testing and the spatial range of the grid. Taking the theoretical coordinates of the embedded parts as a reference, establish a three-dimensional rectangular coordinate system within the detection area. Here, the origin is the geometric center of the embedded parts, the X-axis is along the length direction of the cantilever beam, the Y-axis is along the width direction of the balcony slab, and the Z-axis is along the thickness direction of the welding interface. Specifically, it includes: According to the resolution of the ultrasonic testing equipment (for example, the longitudinal resolution is 0.5 mm and the transverse resolution is 1 mm), combined with the calculation efficiency, select the grid cells as cubes with a side length of 5 mm (this size is an integer multiple of the resolution, which can not only ensure the detection accuracy but also avoid data redundancy). For example, if the range of the detection area in the X-axis direction is 600 mm (1800 - 1200), then the X-axis direction can be divided into 120 grid cells (600 mm ÷ 5 mm / cell). Take the geometric center of a single embedded part as the origin (if there are multiple embedded parts, take the average coordinates of the geometric centers of all embedded parts as the origin). For example, if the coordinates of two embedded parts are (1200, 600, 180) and (1800, 1000, 220), then the X coordinate of the origin is (1200 + 1800) ÷ 2 = 1500 mm, the Y coordinate is (600 + 1000) ÷ 2 = 800 mm, and the Z coordinate is (180 + 220) ÷ 2 = 200 mm, that is, the origin is (1500, 800, 200). Axis definitions: X-axis: Along the length direction of the cantilever beam, starting from the origin, the direction pointing to the cantilever end of the cantilever beam is the positive direction (for example, from the side where the fixed end is connected to the main structure to the side of the balcony slab at the cantilever end).

[0031] Y-axis: Perpendicular to the X-axis, along the width direction of the balcony slab, centered at the origin, extending to both sides of the balcony slab (for example, the left side is the negative direction and the right side is the positive direction).

[0032] Z-axis: Perpendicular to the welding interface, along the thickness direction, centered at the origin, the direction from the base material of the cantilever beam to the base material of the balcony slab is the positive direction (for example, up is positive and down is negative).

[0033] Starting from the origin, evenly divide the grid lines along the three directions of the X-axis, Y-axis, and Z-axis to form a regular cubic grid. Specifically, it includes: Starting from the X coordinate of the origin (such as 1500 mm), draw grid lines at intervals of 5 mm in the positive and negative directions. For example, in the positive direction of the X-axis, they are 1505 mm, 1510 mm... until reaching the maximum range of the X-axis (1800 mm); in the negative direction, they are 1495 mm, 1490 mm... until the minimum range of the X-axis (1200 mm). Each grid line represents a position node in the length direction of the cantilever beam, and the spacing between adjacent grid lines is 5 mm. Division in the Y-axis direction: Taking the origin Y coordinate (such as 800 mm) as the center, divide the grid lines at intervals of 5 mm on both sides of the balcony slab. For example, in the positive Y-axis direction, it is 805 mm, 810 mm... up to the maximum Y coordinate of 1000 mm; in the negative direction, it is 795 mm, 790 mm... down to the minimum Y coordinate of 600 mm, covering the entire detection area in the width direction of the balcony slab. Division in the Z-axis direction: Along the thickness direction of the welding interface, taking the origin Z coordinate (such as 200 mm) as the center, divide the grid lines at intervals of 5 mm upward (towards the balcony slab base material direction) and downward (towards the cantilever beam base material direction). For example, in the positive Z-axis direction, it is 205 mm, 210 mm... up to 250 mm; in the negative direction, it is 195 mm, 190 mm... down to 150 mm, ensuring that the complete thickness (100 mm) of the welding interface is covered.

[0034] The grid lines of the X, Y, and Z axes intersect in space to form a cubic grid. Each intersection point is a grid point, and its coordinates are composed of the grid line coordinates of the three axes (such as X = 1500 mm, Y = 800 mm, Z = 200 mm is the grid point at the origin; X = 1505 mm, Y = 805 mm, Z = 205 mm is an adjacent grid point). All grid points form a regular three-dimensional grid matrix, covering the entire detection area.

[0035] Embed the theoretical geometric features of the welding nodes in the cubic grid to form a reference space grid model containing the coordinates of all grid points and the theoretical geometric features, including the theoretical position of the welding interface, the theoretical axis of the cantilever beam, and the theoretical contour of the embedded parts. Specifically include: According to the design drawings, determine the theoretical range of the welding interface on the Z-axis (such as a thickness of 100 mm, that is, Z = 200 ± 50 mm). In the grid model, mark all grid points with Z coordinates between 150 mm and 250 mm as the "welding interface area", and record the theoretical thickness boundaries of this area (the upper surface Z = 250 mm, the lower surface Z = 150 mm) as the basis for judging whether the subsequent detected defects are located within the welding interface.

[0036] Mark the theoretical axis of the cantilever beam: The theoretical axis of the cantilever beam is a straight line passing through the origin and along the X-axis direction, that is, the set of all grid points that satisfy Y = the origin Y coordinate (such as 800 mm) and Z = the origin Z coordinate (such as 200 mm) (such as X = 1200, Y = 800, Z = 200; X = 1300, Y = 800, Z = 200, etc.). In the grid model, mark these points as the "theoretical axis of the cantilever beam" as the reference line for calculating the lateral offset vector of the defect (such as the distance from the defect center to the axis).

[0037] Embed the theoretical contour of the embedded parts: Rectangular embedded part: If the designed dimensions of the embedded part are 200 mm in length, 100 mm in width, and 80 mm in height (centered at the origin), then the range of its contour in the grid is as follows: X-axis: 1500 - 100 mm to 1500 + 100 mm (1400 mm to 1600 mm); Y-axis: 800 - 50 mm to 800 + 50 mm (750 mm to 850 mm); Z-axis: 200 - 40 mm to 200 + 40 mm (160 mm to 240 mm).

[0038] Mark the grid points within the above range as "embedded part contour".

[0039] Cylindrical embedded part: If the diameter of the embedded part is 60 mm and the height is 100 mm, then centered at the origin, the radius of the bottom circle in the XY plane is 30 mm (i.e., satisfying (X - 1500) 2 +(Y - 800) 2 ≤30 2 ), and the Z-axis range is 200 - 50 mm to 200 + 50 mm (150 mm to 250 mm). Mark the grid points that meet the conditions as "embedded part contour".

[0040] Associate the coordinates (X, Y, Z) of all grid points with their corresponding theoretical geometric features (such as whether they belong to the welding interface, the axis of the cantilever beam, the embedded part contour), and form a reference spatial grid model that includes spatial coordinates and engineering semantics. For example, in addition to recording the coordinates, each grid point is attached with attribute labels (such as "upper surface of the welding interface", "point on the axis of the cantilever beam", "point inside the embedded part", etc.), providing a geometric reference for the subsequent registration of the defect map and the reference model.

[0041] In the embodiment of the present invention, by extracting the theoretical coordinates of the embedded parts in the architectural design drawings (with an accuracy of ±0.5 mm), a three-dimensional rectangular coordinate system is established with the geometric center of the embedded part as the origin, enabling millimeter-level spatial alignment between the detection data and the design reference. Compared with the traditional detection that relies on manual measurement of reference points (with an error of more than ±2 mm), this method eliminates the coordinate system conversion error and ensures the absolute consistency between the defect spatial coordinates and the design drawings. For example, in a certain project, the theoretical coordinates of the embedded part are (1000, 500, 300) mm. During actual detection, the spatial positioning of all ultrasonic signals is based on this origin, avoiding misjudgment of the defect position caused by reference drift. According to the ultrasonic detection resolution (longitudinal 0.5 mm, transverse 1 mm), a 5 mm × 5 mm × 5 mm cubic grid unit is set, which not only meets the sampling density requirements for signal processing but also avoids redundant data. After embedding geometric features such as the theoretical position of the welding interface, the theoretical axis of the cantilever beam, and the theoretical contour of the embedded part, a reference model with engineering semantics is formed: The theoretical position of the welding interface accurately defines the detection range of the fusion zone (such as ±15 mm in the thickness direction) to avoid scanning of invalid areas; the theoretical axis of the cantilever beam, as the core reference line for calculating dynamic offset parameters, is directly related to the structural force analysis (such as the transverse offset vector with the axis as the origin); the theoretical contour of the embedded part automatically identifies the key detection areas (such as the welded section of the anchor bars) to optimize the scanning path planning.

[0042] By mapping the three-dimensional defect distribution map to the coordinates of the reference grid model (error ≤ 0.3 mm), parameters such as the depth gradient of the defect relative to the theoretical axis (such as the expansion rate of cracks per 10 mm depth along the Z-axis) and the transverse offset (such as the center of the air hole deviating 2.5 mm from the Y-axis) can be accurately calculated, providing a quantitative basis for structural mechanics analysis. The grid point coordinates, as a unified data interface, support the subsequent depth residual network to correlate the ultrasonic signal features (such as the amplitude of the stress wave) with the theoretical coordinates (such as the force direction along the X-axis). For example, when identifying unfused defects continuously distributed along the length direction (X-axis) of the cantilever beam, a stress concentration warning can be automatically triggered. The grid division result of the reference model directly drives the robotic arm to scan along the preset path (such as stepping 5 mm along the X-axis and layering 2 mm along the Z-axis), avoiding the scanning blind spots caused by manual operation and increasing the detection efficiency by 40%. The embedded theoretical geometric features, as prior knowledge, assist the time-frequency domain joint threshold segmentation algorithm to exclude non-defect signals (such as pseudo-defect waves generated by reflections from the contour edges of the embedded parts). For example, when a signal with a Z-axis coordinate exceeding the theoretical range of the welding interface thickness (±20 mm) is detected, the system automatically determines it as external clutter of the base material, reducing the misjudgment rate by 60%. By comparing the detected defect coordinates with the theoretical coordinates of the embedded part in real time, a deviation report is automatically generated (such as triggering a red warning when the transverse offset vector of a certain node > 5 mm), and the historical defect coordinate set (including four-dimensional information) is coupled with the structural load data for analysis. For example, through the theoretical axis of the cantilever beam in the reference model, the influence coefficient of the defect on the structural stiffness can be accurately calculated (such as when the stress wave attenuation coefficient increases by 10%, the stiffness decreases by 5%).

[0043] In a preferred embodiment of the present invention, a three-dimensional defect distribution map of the welding interface is generated by the time-of-flight inversion algorithm of the ultrasonic echo signal and spatially registered with the reference grid model to obtain a defect-reference fusion model, including: The probe is attached to the surface of the cantilever beam, emitting high-frequency ultrasonic pulses towards the welding interface and receiving the reflected echo signals. The echo signals at one position are recorded for each scan, including the signal amplitude and the flight time. Specifically, it includes: stably attaching a high-frequency ultrasonic probe (such as a curved wedge probe with a center frequency of 5 - 10 MHz) to the detection surface of the cantilever beam through a magnetic chuck or a robotic arm. The detection area is polished in advance to a roughness Ra ≤ 6.3 μm, and a silicon-based coupling agent (with a thickness of 0.2 - 0.5 mm) is applied to ensure that the sound beam is perpendicular to the welding interface. The probe emits high-frequency ultrasonic pulses (pulse width 0.1 μs) towards the welding interface and synchronously receives the defect reflected echo and the bottom surface echo signals. The A-scan signal at one position is recorded for each scan, including the time-domain waveform (amplitude-time curve), the flight time (the time point corresponding to the peak of the defect echo), and the real-time position coordinates of the probe (obtained through an integrated high-precision positioning module with an accuracy of ±0.1 mm). The "point-by-point stepping and row-column scanning" mode is adopted, with a stepping distance of 2 mm along the length direction (X-axis) of the cantilever beam, covering 20 mm on both sides of the weld seam in the transverse direction (Y-axis), and lifting 1 mm along the thickness direction (Z-axis) after each layer of scanning until the entire thickness of the welding interface (such as a designed thickness of 100 mm) is covered.

[0044] According to the propagation speed of ultrasonic waves in the material, the flight time of the echo signal is converted into the distance between the defect and the probe. Specifically, it includes: measuring the propagation speed of ultrasonic waves in the base material (such as Q345B steel) using a CSK-IA standard test block before detection. At room temperature, the sound speed is approximately 5880 m / s, and it is corrected to an accuracy of ±10 m / s in combination with the on-site temperature (±2 °C). For the flight time t (unit: μs) of the defect wave in the echo signal, the distance d between the defect and the probe is calculated according to the formula d = (v × t) / 2 (v is the sound speed, and it is divided by 2 because the signal propagates back and forth).

[0045] According to the actual position and beam angle during scanning, the coordinates of the defect in the three-dimensional space are calculated. The defect coordinates obtained from each scan are used as a data point, and after multiple scans, a point cloud data containing defect points is formed. Specifically, it includes: real-time recording of the three-dimensional coordinates (Xp, Yp, Zp) of the probe center through a robotic arm encoder or a total station as the scanning position reference. If an oblique-incidence probe (such as a 30° / 60° shear wave probe) is used, the coordinates of the defect in the three-dimensional space are calculated according to the beam incident angle θ: the distance along the sound beam direction is d, and the coordinate Z in the thickness direction perpendicular to the detection surface is Z = Zp + d × cosθ; the lateral offset coordinate X = Xp + d × sinθ × cosϕ; Y = Yp + d × sinθ × sinϕ (ϕ is the deflection angle of the probe in the horizontal plane, and it is along the X-axis direction by default when ϕ = 0°). One defect data point (X, Y, Z) is obtained from each scan. If there is no defect at this position, it is marked as a background point. After multiple scans, a three-dimensional point cloud containing thousands to tens of thousands of points is formed, and the defect points are concentrated in the welding interface area (within the range of ±50 mm along the Z-axis).

[0046] Map the point cloud data into a three-dimensional space grid, divide the cube voxels according to the preset precision, and the gray value of each voxel represents the defect probability of the corresponding area, and finally form a three-dimensional defect distribution map, which specifically includes: taking the coordinate system of the reference grid model as a reference, dividing the detection area into cube voxels with a preset precision (such as 5mm×5mm×5mm), and the voxel size needs to match an integer multiple of the ultrasonic detection resolution (longitudinal 0.5mm, transverse 1mm); map each defect point (X, Y, Z) into the corresponding voxel unit, count the number of defect points in each voxel, if the number of defect points in the voxel exceeds the threshold (such as 3), then mark this voxel as a "suspicious defect area"; according to the defect point density or the average signal amplitude in the voxel, assign a gray value of 0-255 to each voxel (such as the higher the defect probability, the higher the gray value). For example, the gray value of a defect-free voxel is 0, and the gray value of a high-density defect voxel is ≥200. Finally, a three-dimensional gray map containing the spatial distribution of defects is generated, which can be displayed by software three-dimensional visualization (such as the projection of the defect on the X-Y plane, the Z-axis depth slice).

[0047] Extract the four corner points of the embedded part and the edge line of the welding interface from the reference space grid model and the three-dimensional defect distribution map, which specifically includes: directly retrieve the theoretical coordinates of the four corner points of the embedded part from the reference space grid model (such as the corner point coordinates of a cuboid embedded part are (X0±a / 2, Y0±b / 2, Z0±c / 2)), and the theoretical edge line of the welding interface (such as the X-Y plane contour line at Z = Z0±t / 2); perform edge detection on the three-dimensional defect map, and identify the actual corner points of the embedded part (where the signal amplitude changes suddenly) and the edge line of the welding interface (the boundary of the area with dense defect points) through threshold segmentation (such as the gray value ≥150). For example, find the boundary points with a sudden drop in gray value in the map and fit them into a straight line or curve as the actual edge line.

[0048] Match the corner coordinates of the embedded part with the theoretical corner coordinates in the reference grid, calculate the initial translation amount and rotation angle to align their overall positions, calculate the distance from each measured defect point to the nearest theoretical point in the reference grid as the error, and gradually adjust the position and orientation of the three-dimensional defect distribution map until the overall error is less than the set threshold to form a registered defect map. Specifically, it includes: matching the measured corner coordinates of the embedded part with the theoretical corner points of the reference model, calculating the initial translation amount (△X = X measured - X theoretical, △Y, △Z) and rotation angle (Euler angles around the X / Y / Z axes to roughly align the measured corner points with the theoretical corner points in space); using the Iterative Closest Point (ICP) algorithm to calculate the distance from each measured defect point to the nearest theoretical point in the reference grid as the error, and gradually adjusting the position (translation) and orientation (rotation) of the defect map by the least squares method. For example, reduce the error by 10% in each iteration until the overall root mean square error is less than the set threshold (such as 0.5 mm) to ensure that the coordinate deviation between the defect point cloud and the reference grid is within the millimeter range; after registration, randomly select 10 reference grid points to verify the distance from the measured defect points to the corresponding theoretical points. If the error of more than 90% of the points < 1 mm, the registration is considered qualified.

[0049] Embed the registered defect map into the reference grid model so that the coordinates of each defect point correspond to the spatial position of the reference grid, and finally obtain a defect-reference fusion model containing theoretical geometric features and measured defect data. Specifically, it includes: mapping the coordinates of each defect point in the registered defect map to the coordinate system of the reference grid. For example, map the defect point coordinates (1502, 803, 198) to the voxel unit where the nearest grid point (1500, 800, 200) is located. Attach the theoretical geometric attributes in the reference model to each defect point, such as whether it is within the contour of the embedded part (by judging whether the coordinates are within the theoretical dimensions of the embedded part), the distance from the theoretical axis of the cantilever beam, and whether it is within the theoretical thickness range of the welding interface (±50 mm along the Z axis), etc.; merge the grayscale value data (defect probability) of the three-dimensional defect map with the coordinates and theoretical features of the reference grid to form a fusion model containing the measured defect position, spatial distribution density, and theoretical geometric reference.

[0050] In the embodiments of the present invention, the time of flight of the ultrasonic signal (with an accuracy of 0.1 μs) is converted into the defect distance (with an error ≤ 0.3 mm) through the time-of-flight inversion algorithm, and the three-dimensional coordinates are calculated by combining the scanning position (coordinate accuracy ±0.1 mm) and the beam angle (resolution 1°), realizing the millimeter-level positioning of the defect position. Compared with the traditional A / B scan that can only display two-dimensional cross-sectional defects, this method can generate point cloud data containing X / Y / Z axis coordinates, and after voxelization, a three-dimensional map corresponding to the gray value and the defect probability is formed, intuitively presenting the three-dimensional distribution of pores (such as the depth distribution of a spherical pore with a diameter of 5 mm in the Z-axis direction) and the spatial orientation of cracks (such as an unfused crack extending at 45° along the welding interface). In the inspection of a certain bridge precast member, a linear unfused defect with a depth of 12 mm from the surface and a length of 8 mm along the Y-axis direction was successfully located. The traditional method could only judge the existence of the defect and could not accurately describe its spatial form.

[0051] By matching the corner points of the embedded parts (theoretical coordinate accuracy ±0.5 mm) with the measured corner points, the translation amount (accuracy ±0.2 mm) and the rotation angle (±0.5°) are calculated, solving the problem of the inconsistency between the measured data and the design drawing reference in traditional inspections. After registration, the average distance error between the defect points and the theoretical points of the reference grid < 0.8 mm, ensuring the direct correlation between the spatial coordinates of defects such as pores and cracks and the theoretical axis of the cantilever beam and the theoretical position of the welding interface. For example, in the inspection of a welding joint of a balcony slab, after registration of the measured crack starting point coordinates (1502, 805, 195), the lateral offset (Y-axis +5 mm) and depth gradient (Z-axis -5 mm) relative to the theoretical axis can be accurately calculated, providing direct data for evaluating the influence of the defect on the structural force.

[0052] In a preferred embodiment of the present invention, based on the spatial coordinate mapping relationship of the defect-reference fusion model, a deep residual network is used to perform feature fusion analysis on the ultrasonic echo signal to obtain the feature vector corresponding to each signal, including: The original ultrasonic echo signal is obtained. Each signal corresponds to a detection position and contains a curve of the amplitude varying with time. Specifically, it includes: the probe is stably attached to the detection surface of the cantilever beam through a magnetic chuck, and the position control (positioning accuracy ±0.1 mm) is realized in cooperation with a robotic arm or a manual scanning device; each time a scan is triggered, the probe emits a high-frequency ultrasonic pulse (duration 0.1 μs), and the receiving end synchronously collects the echo signal to form A-scan data, that is, a curve of the amplitude varying with time (sampling rate 200 MHz, a single signal contains 1024 time points, and each point records the amplitude voltage value); through the encoder or total station integrated in the device, the three-dimensional coordinates (X, Y, Z) of the current probe center are recorded in real time as the detection position corresponding to this signal, ensuring that each echo signal is bound to the spatial coordinates one by one (such as the coordinates of a certain signal being X = 1500 mm, Y = 800 mm, Z = 200 mm).

[0053] Perform frequency-domain analysis on the ultrasonic echo signal to generate a time-frequency diagram, extract the energy distribution characteristics of the signal in different frequency bands. In the time-frequency diagram, the horizontal axis represents time and the vertical axis represents frequency. Specifically, it includes: dividing the original time-domain signal into overlapping time windows (such as a window length of 20 μs and an overlap rate of 50%), performing Fourier transform on the signal within each window to obtain the frequency components and amplitudes corresponding to different time points; squaring the amplitude at each frequency point to obtain the energy value of that frequency band, forming a two-dimensional time-frequency diagram, with the horizontal axis being time (unit: μs) and the vertical axis being frequency (range: 0 - 20 MHz), and the brightness or color of each pixel representing the energy intensity of the corresponding time-frequency point (for example, the energy concentration area in the high-frequency band of 5 - 10 MHz may indicate lack of fusion defects); extracting the energy statistical characteristics of different frequency bands, such as: The energy ratio in the low-frequency band (0 - 5 MHz), which reflects the overall acoustic characteristics of the base metal matrix; the energy attenuation rate in the medium-high frequency band (5 - 15 MHz), an indicator of high-frequency energy loss caused by defects (such as cracks); the peak frequency offset, compared with the reference frequency of the defect-free signal to determine whether there is an abnormal interface reflection.

[0054] The detection position coordinates corresponding to each signal. Integrate the ultrasonic echo signal, the time-frequency diagram, and the position coordinates into an input tensor. Specifically, it includes: normalizing the original amplitude-time curve (subtracting the mean and dividing by the standard deviation), and unifying the length to 1024 time points by zero-padding or truncation to form a one-dimensional array (shape: 1024,); adjusting the two-dimensional time-frequency diagram to a fixed size (such as 64×64 pixels), normalizing the pixel values to the interval [0, 1] as a two-dimensional matrix input (shape: 64, 64); converting the three-dimensional coordinates (X, Y, Z) of the detection position to floating-point numbers and scaling them according to the coordinate range of the reference grid model (for example, if the X-axis range is 1000 - 2000 mm, then the X coordinate is divided by 2000 and normalized to [0.5, 1]) to form a three-dimensional vector (shape: 3,); combining the time-domain array, the time-frequency diagram matrix, and the position vector into a multi-dimensional input tensor. The specific structure is: Channel dimension, including 1 time-domain channel, 1 time-frequency diagram channel (64, 64), and 1 position channel; the final tensor shape: (1024, 64, 64, 3), corresponding to the multi-modal information fusion of time points, frequency points, and spatial coordinates respectively.

[0055] Generate a comprehensive feature vector containing time-frequency domain signal features and spatial position information based on the input tensor and the trained deep residual network, specifically including: receiving a multi-dimensional tensor and adapting the data of the three channels of time domain, frequency domain, and position; using a 1D convolutional layer (kernel size 5) to extract local features of the time-domain signal (such as echo peak time, rising edge slope); processing the time-frequency map through a 2D convolutional layer (kernel size 3×3) to capture the texture pattern of the energy distribution (such as the linear energy attenuation band of cracks, the circular energy concentration area of pores); mapping the three-dimensional coordinates into a high-dimensional feature vector (such as 64-dimensional) through a fully connected layer and concatenating it with the feature map output by the convolutional layer; introducing skip connections in the deep network to directly transfer the shallow features to the deep layer to avoid gradient disappearance, for example, adding a residual block between the 3rd layer and the 6th layer to maintain the integrity of feature propagation; performing average pooling on the convolutional feature map and merging it with the position feature vector to form a comprehensive feature vector containing multi-modal information (such as 256-dimensional).

[0056] Use the labeled ultrasonic signal data (including pore, lack of fusion, crack, and defect-free samples), divide it into a training set and a validation set according to the ratio of 8:2, adopt cross-entropy loss (for defect classification tasks), select Adam (learning rate 0.001) as the optimizer, iterate for 50 - 100 epochs, monitor the accuracy of the validation set (target > 95%), and avoid overfitting through early stopping. The input tensor undergoes forward propagation through the network, passing through each convolutional block, residual block, and pooling layer in sequence, and finally outputs a feature vector with a fixed dimension (such as 256-dimensional) in the fully connected layer. Each dimension in the feature vector corresponds to different fusion features, for example: Dimensions 1 - 64: Waveform detail features of the time-domain signal (such as the interval between multiple reflected echoes); Dimensions 65 - 192: Energy distribution patterns in the time-frequency map (such as the degree of energy attenuation in a specific frequency band); Dimensions 193 - 256: Spatial position-related features (such as the distance of the defect from the axis of the cantilever beam, the relative position in the embedded part).

[0057] In the embodiment of the present invention, by integrating the original time-domain signal (amplitude-time curve), frequency-domain energy distribution (time-frequency map), and spatial position coordinates (X / Y / Z axes), a three-dimensional input tensor containing signal physical characteristics and structural geometric information is constructed, which solves the problem of information one-sidedness in traditional single-modal analysis (only time domain or frequency domain).

[0058] For example: Time-domain signal: Capturing the amplitude mutation of defect reflection waves (such as multiple reflection echoes caused by cracks); Frequency-domain feature: Identifying the energy attenuation in specific frequency bands (such as the energy loss rate of lack of fusion defects reaching 30% in the 5 - 8 MHz frequency band); Spatial coordinates: Locating defects in the key stress area of the cantilever beam (such as defects within 5 mm from the theoretical axis triggering key warnings); The fusion of the three enables the model to extract composite features such as "cracks at 150 mm from the cantilever end with a sudden drop in energy in the 3 - 6 MHz frequency band", and the recognition accuracy is improved by 25% compared with traditional single-signal analysis. Using Residual Connection to solve the problem of gradient disappearance in deep networks and realizing effective training of deep networks with more than 18 layers, it can capture deep non-linear features in ultrasonic signals: Underlying features: Extracting basic features such as the slope of the rising edge of the signal and the peak time difference through convolutional layers; Middle-level features: Capturing the texture patterns of energy distribution in the time-frequency diagram (such as circular energy concentration areas of pores and linear energy attenuation bands of cracks); High-level features: Combining spatial coordinates to learn the correlation pattern between defect distribution and structural stress (such as the influence weight of lack of fusion defects distributed along the axis of the cantilever beam on stiffness increasing by 40%). Taking the detection position coordinates (accuracy ±0.1 mm) as part of the input tensor, the feature vector contains the "structural semantic" information of the defect: Embedded part association: Automatically identifying defects located within the contour of the embedded part (such as pores at the welded joint of the anchor bar, whose spatial coordinates trigger the corrosion risk model of the embedded part); Axis offset quantification: Calculating the distance between the defect center and the theoretical axis of the cantilever beam through coordinates (such as when the lateral offset vector > 10 mm, the weight of the corresponding dimension in the feature vector increases); Depth gradient encoding: Combining the Z-axis coordinate with the theoretical thickness of the welding interface to distinguish surface defects (Z < 5 mm) from internal defects (Z > 20 mm), and the influence coefficient of the former on the fatigue life is automatically multiplied by a correction factor of 1.5 times.

[0059] Traditional detection relies on engineers to manually extract signal features (such as peak amplitude, number of cycles), which has problems of strong subjectivity and low efficiency. This method realizes full automation from feature extraction to fusion through a data-driven deep learning model: Automatic generation of time-frequency diagrams, generating time-frequency diagrams based on the Short-Time Fourier Transform (STFT) to avoid errors caused by manual selection of window functions; Unifying the dimension of feature vectors, regardless of the signal length, outputting feature vectors with a fixed dimension (such as 256 dimensions) through the pooling layer of the residual network, which is convenient for subsequent threshold segmentation algorithms; Multi-scale feature fusion, capturing multi-scale features of the signal through different convolution kernel sizes (3×3, 5×5), for example, simultaneously identifying the local energy features of 5 mm pores and the global distribution features of 20 mm cracks.

[0060] In a preferred embodiment of the present invention, according to the feature vector and the time-frequency domain joint threshold segmentation algorithm, the types of welding defects are identified and the defect spatial coordinate set is marked to form a defect coordinate set containing four-dimensional information. The types of welding defects include pores, lack of fusion and cracks, including: From the feature vector, key parameters related to the time-frequency domain characteristics of the signal are separated, including time-domain characteristics and frequency-domain characteristics. The time-domain characteristics include peak amplitude, rise time and duration; the frequency-domain characteristics include center frequency, bandwidth, energy centroid frequency and characteristic frequency components, specifically including: In the time-domain waveform of the original ultrasonic echo signal, the maximum voltage amplitude of the defect echo is identified, which reflects the strength of the defect reflection energy (for example, due to strong interface reflection of pores, the peak amplitude is usually more than 30% higher than the base material background signal). Calculate the time required for the signal to rise from 10% to 90% of the peak amplitude to distinguish the defect type (for example, due to the irregular surface of cracks, the rise time is shorter than that of pores, reflecting the fast reflection characteristic of stress waves). Record the time span of the defect echo signal from starting to exceed the noise threshold to falling below the threshold. Due to the large interface area of lack of fusion defects, the duration is usually 2-3 times longer than that of pores.

[0061] Frequency-domain feature extraction: Center frequency: Perform Fourier transform on the signal and calculate the weighted average of the frequency-energy distribution (the frequency band with higher energy contributes more to the center frequency). Due to the high-frequency energy attenuation of lack of fusion defects, the center frequency is often 5-10% lower than the base material reference value.

[0062] Bandwidth: Defined as the frequency interval width where the energy accounts for 90% of the total energy. Due to multi-scale interface reflection of crack defects, the bandwidth is 15-20% wider than that of pores.

[0063] Energy centroid frequency: Determine the core frequency point of the energy distribution by integrating the product of frequency and corresponding energy and normalizing (for example, the energy centroid of pores is concentrated at 8-10 MHz, while due to the scattering effect of cracks, the centroid shifts to 5-7 MHz).

[0064] Characteristic frequency components: Identify the energy anomaly in a specific frequency band (for example, a sudden drop in energy at 5 MHz may indicate the presence of a lack of fusion interface). Combine the welding process parameters in the design drawing (such as the acoustic impedance matching frequency of the base material) to screen the sensitive frequency band.

[0065] Integrate the time-domain characteristics and frequency-domain characteristics of each detection signal into a matrix. Each row of the matrix corresponds to the signal feature vector at a detection position, including the time-domain and frequency-domain parameters of the signal; each column of the matrix corresponds to the specific feature parameter, forming a multi-dimensional feature space, specifically including: Create a row of data for each detection location (uniquely identified by the X / Y / Z coordinates recorded by an encoder or total station), including all time-domain and frequency-domain characteristic parameters of the signal at that location (such as peak amplitude, rise time, center frequency, etc., a total of 7-10 characteristics); the column names of the matrix correspond to specific characteristics (such as "Time Domain - Peak Amplitude", "Frequency Domain - Center Frequency") to form a structured table, and the data in each column is standardized (such as normalized to the interval [0, 1]) to eliminate the influence of dimensions; the characteristic matrix is horizontally extended into a two-dimensional table of "Detection Location × Characteristic Parameter", and each cell stores the characteristic value corresponding to the location (such as the peak amplitude of the location (1500, 800, 200) is 85 mV, and the center frequency is 7.2 MHz); observe the distribution clustering of different defect types in the feature space through a visualization tool (such as a heat map) (such as pores concentrated in the "high peak - narrow frequency band" area, and cracks distributed in the "fast rise edge - low energy center of gravity" area) to assist in verifying the effectiveness of feature separation.

[0066] Based on the training of historical detection data and defect samples, for three types of defects, namely pores, lack of fusion, and cracks, jointly determine the time-domain and frequency-domain characteristic judgment conditions, specifically including: Collect at least 1000 labeled samples (including pores, lack of fusion, cracks, and defect-free signals), and divide them into a training set and a validation set according to 8:2; conduct statistical analysis on the characteristic parameters of each type of defect, and calculate the mean, standard deviation, and distribution interval (such as the mean peak amplitude of pores is 90 mV, and the standard deviation is 10 mV; the mean duration of lack of fusion is 12 μs, and the standard deviation is 3 μs); condition formulation: For pores, set "peak amplitude > 80 mV and frequency band width < 3 MHz and center frequency of energy center of gravity > 8 MHz" (high reflection in the time domain, narrow-band high energy in the frequency domain).

[0067] For lack of fusion, combine "duration > 10 μs and center frequency < 7 MHz and energy attenuation of characteristic frequency component (5 MHz) > 20%" (long duration in the time domain, energy loss in the low-frequency band in the frequency domain).

[0068] For cracks, define "rise time < 0.8 μs and frequency band width > 5 MHz and periodic oscillation of characteristic frequency component (6 MHz)" (fast rise in the time domain, wide-frequency scattering in the frequency domain).

[0069] The conditions are combined through logical "AND" and "OR" to avoid misjudgment by a single characteristic (such as only a high peak may be caused by bubbles in the surface coupling agent, and it is necessary to simultaneously meet the narrow frequency band to be determined as a pore).

[0070] According to the characteristic parameters of each detection signal and the joint judgment conditions, judge each feature vector of the signals in the matrix one by one to obtain the defect signals and the corresponding detection location coordinates of the defect signals, specifically including: For each row in the feature matrix (i.e., the signal at each detection position), sequentially check whether the time-domain and frequency-domain features meet the joint determination conditions for the corresponding defects (e.g., first determine whether it meets the porosity condition, then lack of fusion, and finally crack, ensuring mutual exclusivity). Example: A signal has a peak amplitude of 95 mV (>80 mV), a frequency band width of 2.5 MHz (<3 MHz), and an energy centroid frequency of 8.5 MHz (>8 MHz), meeting the porosity condition. Mark it as a porosity defect and record its detection position coordinates (X = 1505, Y = 802, Z = 198).

[0071] If the signal features simultaneously meet the conditions of multiple defects, it is preferentially determined as the high-risk type (e.g., crack > lack of fusion > porosity); if none of them are met, mark it as "no defect" or "suspicious signal" (triggering manual review), and combine the geometric constraints of the reference model (e.g., when the Z-axis coordinate exceeds the theoretical thickness range of the welding interface by ±50 mm, directly exclude the possibility of defects) to reduce misjudgment caused by boundary reflection.

[0072] According to the spatial coordinate mapping relationship of the defect-reference fusion model, convert the detection position coordinates corresponding to the defect signal into three-dimensional coordinates in the defect-reference fusion model, and add the corresponding welding defect type label to each defect coordinate to form a set of defect coordinates containing four-dimensional information; the defect-reference fusion model includes the theoretical geometric features of the welding node and the measured defect data, that is, the set of defect coordinates. The theoretical geometric features include the theoretical axis of the cantilever beam and the theoretical position of the welding interface, specifically including: According to the coordinate system of the defect-reference fusion model (with the geometric center of the embedded part as the origin, and the X / Y / Z axes corresponding to the length of the cantilever beam, the width of the balcony slab, and the thickness of the welding interface respectively), convert the original coordinates recorded by the detection device (such as the probe center coordinates) into three-dimensional coordinates in the reference model through translation and rotation (e.g., the original coordinates (1505, 802, 198) are converted into reference coordinates (5, 2, -2), unit mm, with the origin as the reference point). Randomly select 10% of the defect coordinates and compare them with the theoretical coordinates of the reference model through total station measurement to ensure that the deviation <0.5 mm (e.g., the lateral offset of a crack coordinate after conversion from the theoretical axis is 3 mm, recorded as Y-axis +3 mm).

[0073] Assign type labels ("porosity", "lack of fusion", "crack") to each defect coordinate, and record the detection timestamp (accurate to the millisecond level, such as 2025-04-14 10:05:30.123); each defect entry contains "X / Y / Z coordinates + timestamp + defect type" (such as (1505, 802, 198, 20250414100530123, porosity)), and is associated with the theoretical features in the reference model (such as whether it is within the profile of the embedded part, the distance from the theoretical axis, etc.). Store the set of defect coordinates as a structured file (such as CSV or JSON), containing all four-dimensional information and additional attributes (such as the detection device number, environmental temperature and humidity), and support importing into the BIM system or 3D visualization software to generate a digital twin model of the welded joint with defect labels.

[0074] In the embodiments of the present invention, the time-domain features (peak amplitude, rise time, duration) and frequency-domain features (center frequency, bandwidth, energy centroid frequency) are separated to construct a multi-dimensional feature space that includes the dynamic waveform and frequency energy distribution of the signal. For example: Porosity: In the time domain, it shows isolated spikes (high peak value and short duration), and in the frequency domain, it presents narrowband high-energy concentration; Lack of fusion: In the time domain, it is accompanied by multiple reflected waves (the duration increases), and in the frequency domain, the energy significantly attenuates in the 5-8 MHz frequency band; Crack: In the time domain, it shows periodic oscillations (steep rise edge), and in the frequency domain, the characteristic frequency components shift.

[0075] Compared with traditional single time-domain or frequency-domain analysis, the combined features can capture the composite physical properties of defects. The recognition accuracy of the three types of defects is increased by more than 30%, and the recognition rate of micro-cracks less than 2 mm is increased from 60% to 92%. Train the combined judgment conditions through historical data (such as "reflection energy mutation > 20 dB and phase shift > 15°" for lack of fusion defects), replace manual experience judgment, and avoid missed judgments caused by subjective differences of operators (such as misjudging the base metal oxide film as lack of fusion), and the misjudgment rate is reduced by 60%. Convert the detection signal coordinates into three-dimensional coordinates in the defect-reference fusion model (error ≤ 0.8 mm), and associate geometric features such as the theoretical axis of the cantilever beam and the theoretical position of the welding interface. For example: The specific coordinates of the crack on the X / Y / Z axis (such as 1502, 805, 195) and the angle with the theoretical axis (45°) are marked to intuitively present the spatial distribution of defects inside the welding node (such as cracks extending along the force direction); the three-dimensional volume of the pores (calculated by the density of the defect points of adjacent voxels) and the interface separation area of ​​the unfused defects are quantified to provide direct input parameters for structural mechanics analysis. The traditional method can only provide the defect position of the two-dimensional section. This method realizes the three-dimensional presentation of the defect space morphology to avoid safety hazards caused by misjudgment of the position (such as missing microcracks in the core stress area). Based on the theoretical axis of the benchmark model, the defect depth gradient (such as the crack extends 0.5mm for every 10mm along the Z axis) and the lateral offset vector (such as the center of the pore deviates from the axis by 2.5mm) are automatically calculated, which directly relates to the structural force analysis. In a certain engineering example, the stress wave attenuation coefficient exceeded the standard (>threshold 10%), and the node stiffness was confirmed to decrease by 15% in combination with finite element simulation, providing data support for timely reinforcement.

[0076] The defect coordinate set includes the detection timestamp (accurate to milliseconds), forming a four-dimensional data chain of "time-space-type-parameter", which supports the comparison of historical detection data (such as the expansion trend of defects at the same location over time). For example: The dense pore defect at a node was laterally offset by 5mm during the inspection in 2023, and the offset increased to 8mm during the retest in 2025, triggering the update of the structural life prediction model, and the predicted remaining life was adjusted from 22 years to 18 years, realizing dynamic monitoring of defect development. The defect coordinates and type labels are connected to the building information model (BIM), and the component design parameters (such as the theoretical outline of embedded parts), construction records (such as welding process parameters), and operation and maintenance data (such as load monitoring values) are associated to form a full-chain quality traceability system. After application in a certain affordable housing project, the efficiency of welding node defect rectification increased by 40%, and the cost of later structural monitoring decreased by 25%, realizing closed-loop management from detection to maintenance.

[0077] The deep residual network automatically extracts 256-dimensional feature vectors, replacing manual calculations of peak values, frequencies and other parameters (traditionally requiring 5-10 minutes per signal). The single-node detection time is shortened from 30 minutes to 5 minutes, and 16-channel synchronous acquisition is supported, making it suitable for rapid detection of batch prefabricated components in prefabricated buildings. Based on the preset joint judgment conditions of the training data (such as "high-frequency energy attenuation rate > 25% and duration > 5μs is judged as a crack"), automatic classification of defect types is achieved, avoiding differences in detection results caused by inconsistent threshold settings in traditional methods, and improving the standardization of the detection process by more than 90%.

[0078] Through joint analysis in the time and frequency domains, the bottom surface echo noise is suppressed (the recognition rate is still maintained at 85% when the signal-to-noise ratio is ≥15dB). For example, when the surface roughness Ra≤6.3μm is detected, the algorithm can effectively distinguish the echo difference caused by insufficient welding current (local lack of fusion) and the parent material oxide film (functional interface) under the guarantee of the uniformity of the silicon-based coupling agent, avoiding the misjudgment of false defects; for the oblique incidence scanning (30° / 60° shear wave) of the complex area of ​​the fillet weld, combined with the embedded part contour annotation of the benchmark model, the pseudo-defect waves generated by the geometric boundary reflection are automatically filtered, and the missed detection rate is reduced from 12% to 3%.

[0079] In a preferred embodiment of the present invention, the dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam are calculated in the defect-reference fusion model, and the parameters include the defect depth gradient, the lateral offset vector and the stress wave attenuation coefficient, including: Extract the information of the theoretical axis of the cantilever beam from the defect-benchmark fusion model, including the direction, position and related geometric parameters of the axis; determine the defect coordinate set, each defect coordinate contains its position information in three-dimensional space, specifically including: retrieve the geometric center coordinates of the embedded parts from the benchmark space grid model as the origin (such as (1500, 800, 200)), the theoretical axis of the cantilever beam is defined as a straight line passing through the origin and along the X-axis direction (all points that satisfy Y=800mm, Z=200mm); the positive direction of the X-axis points to the cantilever end of the cantilever beam (the direction of the balcony slab), and the negative direction points to the fixed end (the main structure connection side), and the axis position is uniquely determined by the origin coordinates and the X-axis extension range (such as X-axis 1200-1800mm); the geometry The parameters include the spatial direction of the axis (linearly extending along the X-axis), the reference point (origin) and the relative position to the welding interface (the axis is located at the center of the theoretical thickness of the welding interface, Z = 200mm); all coordinate points marked as "pores", "unfused" and "cracks" are extracted from the defect-reference fusion model. Each coordinate contains three-dimensional position information (X, Y, Z). For example, the coordinates of a crack are (1502, 805, 195), which means that it is 1502mm in the length direction of the cantilever beam, 5mm to the right of the balcony slab width direction, and 5mm below the welding interface thickness direction; abnormal coordinate points are eliminated (such as points where the Z axis exceeds the theoretical thickness range of the welding interface by ±50mm, which are determined to be interference from external clutter of the parent material) to ensure that only coordinate data within the effective defect area is retained.

[0080] According to the coordinate system and geometric features in the defect-reference fusion model, determine the depth direction, which is along the direction perpendicular to the surface of the cantilever beam and pointing towards the inside of the weld. Specifically, it includes: According to the definition of the reference model coordinate system, the surface of the cantilever beam (detection surface) is the X-Y plane (for example, Z = 250 mm is the surface of the balcony slab base material, and Z = 150 mm is the surface of the cantilever beam base material). The depth direction is perpendicular to this surface, that is, along the Z-axis direction; the depth direction points towards the inside of the weld interface, specifically extending from the surface (Z-axis boundary) towards the central axis (Z = 200 mm). For example: from the surface of the balcony slab (Z = 250 mm) towards the negative Z-axis direction (cantilever beam base material), or from the surface of the cantilever beam (Z = 150 mm) towards the positive Z-axis direction (balcony slab base material), and finally pointing towards the core area of the weld interface; verify the depth direction through the theoretical position of the weld interface in the reference model (such as the range of Z-axis from 150 - 250 mm), ensure that the Z-axis values of all defect coordinates are within this range, and the depth increasing direction is consistent with the positive Z-axis direction (positive for the balcony slab base material direction and negative for the cantilever beam base material direction).

[0081] Group the set of defect coordinates according to their positions in the length direction of the cantilever beam. For the defects within each group, calculate their coordinate values in the depth direction, specifically including: Divide the length direction of the cantilever beam (X-axis) into continuous intervals at a preset interval (such as 5 mm). For example, the X-axis range from 1200 - 1205 mm is the 1st group, 1205 - 1210 mm is the 2nd group, until covering the X-axis range of all defect coordinates (such as 1200 - 1800 mm); for each defect coordinate, classify it into the corresponding group according to its X value. For example, X = 1502 mm is classified into the 1500 - 1505 mm group, and X = 1506 mm is classified into the 1505 - 1510 mm group. Allow overlapping boundaries between groups (such as including the left endpoint and not including the right endpoint) to avoid duplicate classification. For all defects within each group, extract their Z-axis coordinate values (depth direction positions). If there are multiple defects within the same group (such as the coordinates of 3 pores are Z = 190, 195, 200 mm respectively), you can choose to calculate the average value (195 mm) or retain all values for subsequent gradient analysis (such as calculating the maximum and minimum depth changes). If there are no defects within the group (that is, no detected defects in this length interval), then skip this group or mark it as a "defect-free group" and do not participate in the subsequent gradient calculation.

[0082] For the defects in adjacent groups, calculate the ratio of the coordinate difference in the depth direction to the distance difference in the length direction of the cantilever beam, that is, the defect depth gradient. Specifically, it includes: Arrange all defective groups in the order of the X-axis (e.g., the X range of the nth group is 1500 - 1505 mm, and the X range of the (n + 1)th group is 1505 - 1510 mm), ensuring that adjacent groups are continuous in the length direction without gaps. Extract the representative X value of each group (e.g., the midpoint of the X-axis within the group, taking 1502.5 mm for the nth group and 1507.5 mm for the (n + 1)th group) as the position identifier of the group in the length direction.

[0083] For each pair of adjacent groups, obtain their coordinate values in the depth direction (e.g., the average Z of the nth group is 195 mm, and the average Z of the (n + 1)th group is 205 mm), calculate the depth difference (205 - 195 = 10 mm) and the distance difference in the length direction (1507.5 - 1502.5 = 5 mm); the gradient value is the depth difference divided by the distance difference (10 mm / 5 mm = 2 mm / mm), indicating that for every 1 mm of extension along the length direction of the cantilever beam, the defect depth increases by 2 mm in the positive Z-axis direction (the base material of the balcony slab), reflecting the extension trend of the defect in the thickness direction.

[0084] If the depth difference between adjacent groups is negative (e.g., the Z of the latter group is less than that of the former group), the gradient is negative, indicating that the defect depth extends in the negative Z-axis direction (the base material of the cantilever beam), and it is necessary to combine the theoretical position of the welding interface to judge whether it points to the core stress area (e.g., near Z = 200 mm); if the distance between adjacent groups is too large (e.g., exceeding 10 mm) or the number of defects is too small (less than 2 in a single group), then ignore the gradient calculation of this group pair to avoid errors caused by sparse data; if the standard deviation of the depth coordinates of a certain group > 5 mm (indicating a high dispersion of defect depths within the group), then use the median instead of the average value to reduce the influence of outliers.

[0085] In the embodiments of the present invention, by calculating the change rate of the defect depth (Z-axis coordinate) in groups along the length direction (X-axis) of the cantilever beam, the extension trend of the defect in the thickness direction is visually presented. For example: for a certain crack in the range of 1500 - 1510 mm on the X-axis, the Z-axis coordinate increases from 190 mm (close to the surface of the cantilever beam base material) to 210 mm (extending into the base material of the balcony slab), and the depth gradient is (210 - 190) / (1510 - 1500) = 2 mm / 1 mm, indicating that the crack spreads towards the core stress-bearing area at a rate of 2 mm per millimeter of length, triggering a high-risk warning (the traditional method can only detect single-depth defects and cannot identify the extension trend); if the depth gradient of pore-like defects is close to 0 (such as at the same X-axis position, the Z-axis coordinate fluctuation < 1 mm), it is determined as a local isolated defect with relatively low harmfulness; conversely, if the depth gradient > 1 mm / 1 mm (such as the lack of fusion defect rapidly extending along the thickness direction), it indicates a risk of large-area separation at the welding interface. The depth gradient is directly related to the stress distribution in the thickness direction of the structure. For example, the core stress-bearing area of the cantilever beam is usually located in the middle of the thickness direction (near the theoretical axis of the Z-axis). If the defect depth gradient points to this area (such as the gradient is positive and the Z-axis coordinate increases towards the core area), the influence coefficient on the bending stiffness of the structure automatically increases by 50%, avoiding the deviation in bearing capacity estimation caused by misjudgment of the defect position.

[0086] Taking the theoretical axis of the cantilever beam (a straight line passing through the origin and along the X-axis, Y = origin Y, Z = origin Z) as the reference, calculate the perpendicular distance (lateral offset) from the defect center to the axis and the offset direction (positive or negative value of the Y-axis). For example: the center coordinates of a lack of fusion defect are (1600, 810, 200), and the theoretical axis coordinates are (1600, 800, 200), then the lateral offset vector is +10 mm on the Y-axis, indicating that the defect is 10 mm to the right of the axis. Combining with the theory of mechanics of materials, for every 5 mm increase in the offset, the bending stress concentration coefficient at this position increases by 15%, directly guiding the parameter input of the finite element simulation model (the traditional method relies on manual measurement of the axis, with an error > 2 mm, resulting in a stress calculation deviation exceeding 20%). In batch detection, if the lateral offset vectors of multiple defects at the same node are all > 5 mm and deviate to the same side (such as the positive direction of the Y-axis), the system automatically determines it as "welding eccentricity" and triggers the traceability of the construction process (such as the positioning deviation of the welding machine), reducing structural hidden dangers from the source. Set the risk level according to the offset (such as < 5 mm is green, 5 - 10 mm is yellow, > 10 mm is red). In an actual project, the lateral offset vector of a crack at the welding joint of a balcony slab reaches 12 mm (exceeding the safety threshold of 10 mm of the theoretical axis). Combining with finite element analysis, it is confirmed that the stress amplitude at this position increases by 30% compared with the design value, predicting the fatigue failure risk 3 years in advance and avoiding the lag of traditional empirical evaluation.

[0087] By comparing the attenuation degree of the stress wave amplitude between the defective area and the defect-free area, the attenuation coefficient is calculated (e.g., attenuation coefficient = 1 - signal amplitude in the defect area / signal amplitude in the reference area), which directly reflects the impact of defects on the vibration transmission efficiency of the structure. For example: Due to the significant scattering effect of dense pore defects, the attenuation coefficient can reach 40% (the amplitude in the defect-free area is 100 mV, and the amplitude in the defective area is only 60 mV). Verified by finite element simulation, the stiffness of this node decreases by 15% compared with the design value, which is consistent with the results of traditional loading tests (the traditional method requires offline tests and takes 1 - 2 weeks, while this method calculates in real-time online and takes <1 minute). Lack of fusion defects cause sudden changes in wave impedance due to interface separation, and the attenuation coefficient is often >30%, accompanied by a phase shift (>10°), which can effectively distinguish between poor welding process (real defect) and uneven base material quality (non-defect signal), reducing the misjudgment rate by 70%. As the core input parameter, the attenuation coefficient drives the life assessment model to learn the correlation law between defect parameters and the structural failure time.

[0088] In a certain project, the attenuation coefficient of a certain node is 25%, and the model predicts the remaining life of 22 years (estimated 30 years by traditional accelerated tests). After 5 years of monitoring, the annual degradation rate of the node stiffness coincides with the predicted curve, providing a reliable time window for operation and maintenance decisions (such as formulating a reinforcement plan 15 years in advance). The combination of depth gradient, lateral offset vector, and attenuation coefficient forms a three-dimensional evaluation system of "position - shape - performance". For example: A certain crack defect simultaneously meets the conditions of "depth gradient of 1.5 mm / mm (extending towards the core area), lateral offset vector of 8 mm (stress concentration area), and attenuation coefficient of 35% (significant stiffness reduction)". The system automatically determines it as "extremely high risk" and gives priority to triggering the emergency reinforcement process; if a single pore defect has "depth gradient of 0, lateral offset of 3 mm, and attenuation coefficient of 10%", it is determined as low risk and can be included in routine monitoring. This multi-parameter joint evaluation avoids the one-sidedness of a single index, increasing the accuracy rate of defect hazard judgment from 60% of the traditional method to 92%; the automatic calculation of dynamic parameters (without manual intervention) is deeply integrated with the geometric features of the reference model (such as the theoretical axis and the thickness of the welding interface), shortening the single-node detection time from 30 minutes to 5 minutes, and simultaneously outputting a multi-dimensional report including spatial distribution, extension trend, and mechanical impact.

[0089] The dynamic parameters, as the core attributes of the defect coordinate set (four-dimensional information), are seamlessly integrated into the BIM system, forming a data closed-loop with design drawings (theoretical coordinates of embedded parts), construction data (welding process parameters), and operation and maintenance monitoring (load history). For example, the historical data of the lateral offset vector of a certain node shows that it has increased from 5 mm to 12 mm in 3 years, and the attenuation coefficient has increased synchronously (from 15% to 30%). The system automatically generates a "defect expansion - stiffness degradation" correlation curve, providing real-time input for the structural health monitoring system and realizing the full-chain digital management from detection to prediction. Based on the statistical analysis of dynamic parameters (such as the average depth gradient of components in the same batch > 0.5 mm / mm), welding process problems in the prefabrication link (such as the deviation of the welding torch angle) can be traced, optimizing the production process from the source and reducing the later operation and maintenance costs by more than 25%.

[0090] In a preferred embodiment of the present invention, the determination process of the lateral offset vector is as follows: According to the coordinate system in the defect-reference fusion model and the geometric information of the cantilever beam, the lateral direction is determined. The lateral direction is the direction perpendicular to the theoretical axis of the cantilever beam and within the plane of the cantilever beam, specifically including: Retrieve the reference origin of the theoretical axis of the cantilever beam from the defect-reference fusion model (such as the geometric center coordinates of the embedded part, for example: (1500, 800, 200)). This axis extends along the X-axis direction (the length direction of the cantilever beam), and all points on the axis satisfy Y = 800 mm and Z = 200 mm (the Y / Z coordinates of the origin). Define the plane where the cantilever beam is located as the X-Y plane (parallel to the length and width directions of the cantilever beam, perpendicular to the Z-axis of the thickness direction). The lateral direction is the direction perpendicular to the theoretical axis (X-axis) and within the X-Y plane, that is, the Y-axis direction in the coordinate system (the positive direction of the Y-axis is the right side of the balcony slab, and the negative direction is the left side). The lateral offset only reflects the position deviation of the defect in the width direction (Y-axis) of the balcony slab and has nothing to do with the thickness direction (Z-axis) (the Z-axis is used for the calculation of the depth gradient), ensuring that the lateral direction is strictly limited within the X-Y plane and avoiding confusion with the depth direction.

[0091] For each defect point in the defect coordinate set, project it onto the theoretical axis of the cantilever beam to obtain the corresponding projection point, specifically including: for any defect point coordinates (Xd, Yd, Zd), the coordinates of the point on the theoretical axis corresponding to the X value of Xd are (Xd, Y-axis, Z-axis), where the Y-axis and Z-axis take the fixed values of the theoretical axis in the reference model (such as Y-axis = 800 mm, Z-axis = 200 mm, the same as the origin). Replace the Y coordinate and Z coordinate of the defect point with the Y-axis and Z-axis of the theoretical axis, and keep the X coordinate unchanged to obtain the projection point coordinates.

[0092] Example: The coordinates of the defect point are (1600, 810, 195), and the point projected onto the theoretical axis is (1600, 800, 200) (the Y-axis is adjusted from 810 mm to 800 mm, and the Z-axis is adjusted from 195 mm to 200 mm. However, according to the definition of the transverse direction, the Z-axis adjustment is only used to locate the axis, and the transverse offset only focuses on the Y-axis difference). Calculate the vector between each defect point and its projected point, that is, the transverse offset vector of the defect point relative to the theoretical axis of the cantilever beam, which contains the magnitude and direction information of the offset. Specifically, based on the theoretical axis, the positive direction of the Y-axis is the right side of the balcony slab, and the negative direction is the left side. If the Y coordinate Yd of the defect point > Y axis (such as 810 mm > 800 mm), the offset direction is the positive direction of the Y-axis (right side); if Yd < Y axis, it is the negative direction of the Y-axis (left side); calculate the absolute distance between the defect point and the projected point in the Y-axis direction, with the unit of millimeters (such as 10 mm means the defect point deviates 10 mm to the right of the axis).

[0093] The transverse offset vector of each defect point is expressed as (offset direction, offset distance). For example: The offset vector of the defect point (1600, 810, 195) is (positive direction of the Y-axis, 10 mm); The offset vector of the defect point (1550, 790, 210) is (negative direction of the Y-axis, 10 mm).

[0094] If the Z coordinate of the defect point exceeds the theoretical thickness range of the welding interface (such as Z < 150 mm or Z > 250 mm), it is determined as an invalid point and does not participate in the offset vector calculation (to avoid interference from external clutter of the base material); for the defect point located on the theoretical axis (Yd = Y axis), the offset vector is marked as (no offset, 0 mm).

[0095] In the embodiment of the present invention, based on the theoretical axis of the cantilever beam (a straight line passing through the origin and along the X-axis), the defect point is vertically projected onto the axis, and the vector offset (such as the deviation in the Y-axis direction) is calculated with an accuracy of ±0.5 mm. For example: After the coordinates of a crack defect (1600, 810, 200) are projected onto the axis, the transverse offset vector is +10 mm in the Y-axis, indicating that the defect is located 10 mm to the right of the axis. According to the principle of material mechanics, for every 5 mm increase in the offset, the bending stress concentration coefficient at this position increases by about 15%, directly providing accurate input parameters for finite element simulation (the error of traditional manual measurement > 2 mm, resulting in a stress calculation deviation exceeding 20%).

[0096] In batch inspection, if the lateral offset vectors of multiple defects at the same node are all biased towards the positive direction of the Y axis (such as the right side of the balcony slab), the system automatically determines it as "welding eccentricity", prompting welding gun positioning deviation or fixture error, reducing structural hidden dangers from the source of construction. Combined with the offset direction and size, a defect distribution heat map is generated (such as the red area indicates a high-risk area with an offset of >10mm). In the inspection of prefabricated components, the lateral offset vector is used to locate the unfused defect 12mm away from the axis. After load test verification, the stress amplitude at this position is 30% higher than the design value, which warns of fatigue failure risks in advance and avoids the lag of traditional empirical evaluation.

[0097] The direction of the lateral offset vector (positive and negative values ​​of the Y axis) reflects the offset trend of the defect in the width direction of the balcony slab. For example, if a large number of defects are concentrated in the positive direction of the Y axis (the right side of the balcony slab), it can be traced back to the workpiece clamping deviation during welding (such as the fixture tilted to the right by 1°), or the welding gun movement path planning error (offset theoretical trajectory 5mm), guiding the construction unit to adjust the process parameters in a targeted manner to avoid the recurrence of similar problems. If the lateral offset vector of the defects around the cylindrical embedded parts is distributed radially (such as the center of the embedded part as the origin and offset to the surroundings), it indicates that the positioning accuracy of the anchor bar is insufficient during welding (such as the verticality deviation of the anchor bar>2°), triggering the calibration process of the automated equipment in the prefabrication link. Defects located near the axis (offset <5mm) have little effect on the uniformity of the structural force, while defects with an offset of >10mm (especially towards the tensile side of the cantilever beam) will significantly change the section moment of inertia, resulting in a decrease in bearing capacity. In an actual project, three high-offset defects in key stress areas were identified through the lateral offset vector, and reinforcement was arranged as a priority, which increased the node rectification efficiency by 40%.

[0098] As the core attribute of the defect space coordinates, the lateral offset vector directly participates in the structural force calculation. For example, when calculating the additional bending moment at the defect (M = F × d, where d is the offset distance and F is the design load), if the offset is 10mm, the additional bending moment increases by 10% compared with the axis, which significantly affects the input parameters of the node stiffness degradation model. Combined with the depth gradient (extension trend in the Z-axis direction), a three-dimensional risk assessment system of "plane offset + thickness extension" is formed. For example, if a crack has both a Y-axis + 8mm offset and a Z-axis + 15mm depth (extending to the core stress area), the system automatically determines that its influence weight on the structural bending strength is twice that of an ordinary defect, avoiding the one-sidedness of a single-dimensional assessment. Using the rigid reference of the theoretical axis (position accuracy ± 0.5mm), the pseudo-defect offset caused by the positioning error of the detection equipment (such as the systematic deviation caused by the tilt of the sensor installation) is filtered. By comparing the distance between the defect coordinates and the axis, abnormal points that exceed the theoretical width of the welding interface (such as the balcony slab design width ± 20mm) are automatically eliminated, and the misjudgment rate is reduced by 60%.

[0099] In a preferred embodiment of the present invention, the process of determining the stress wave attenuation coefficient is: Using the principle of ultrasonic testing and related physical models, simulate the propagation process of stress waves in the cantilever beam and welded joints. Measure the intensity of stress waves at each defect position in the defect coordinate set and at the reference positions on the cantilever beam, specifically including: In the defect-free area of the base metal of the cantilever beam (such as the fixed end far from the welding interface, the theoretical coordinates are verified by the reference model as non-welding areas, for example: X = 1000 mm, Y = 800 mm, Z = 200 mm), ensure that there are no defects at this position and the material is uniform, serving as the reference point for stress wave propagation; Extract all valid defect points from the defect coordinate set (abnormal points outside the welding interface have been excluded, such as the coordinates of pores, lack of fusion, and cracks within the range of 150 - 250 mm on the Z-axis). For example, the coordinates of a certain crack are (1502, 805, 195). Use the same high-frequency piezoelectric ceramic probe as in the signal acquisition stage (center frequency 5 - 10 MHz, arc-shaped wedge matching surface), attach the detection surface at the reference position and the defect position respectively, apply a uniform silicon-based coupling agent (thickness 0.2 - 0.5 mm), and ensure that the sound beam is vertically incident (record the angle parameter when obliquely incident).

[0100] Emit ultrasonic pulses with the same parameters (such as emission voltage 150 V, pulse width 0.1 μs), receive the stress wave echo signals, and focus on recording the peak amplitude of the defect echo (reflecting the stress wave intensity). Collect 3 signals at each position and take the average to reduce accidental errors (such as the three amplitudes at the reference position are 100 mV, 98 mV, 102 mV, with an average of 100 mV; the three amplitudes at the defect position are 70 mV, 72 mV, 68 mV, with an average of 70 mV). Signal calibration and environmental synchronization: Before collection, use a CSK-IA standard test block to calibrate the equipment gain to ensure that the bottom echo amplitude at the reference position is stable at 80% of the full screen; Synchronously record the detection environment parameters (temperature 25 ± 2 °C, humidity 40 ± 5%RH). Since the temperature change affects the sound speed, it is necessary to ensure that the detection environments at the reference and defect positions are the same to avoid intensity measurement deviations caused by environmental differences.

[0101] According to the change of stress wave intensity during propagation, calculate the stress wave attenuation coefficient, specifically including: Extract the average signal amplitude at the reference position (denoted as A_reference, such as 100 mV) and the average signal amplitude at the defect position (denoted as A_defect, such as 70 mV). Both are the peak voltage values in the time domain waveform, reflecting the energy intensity of the stress wave when it propagates to this position. Compare the signal intensity differences between the defect position and the reference position. If A_defect < A_reference, it indicates that the stress wave attenuates due to the presence of defects during propagation (such as energy scattering or reflection by cracks, pores); If the two are close (difference < 5%), it is determined that there is no significant attenuation (possibly small-sized defects or noise signals).

[0102] Qualitatively describe the attenuation coefficient according to the direction and degree of intensity change: No attenuation: The amplitude difference between the defect position and the reference position is less than 5%, marked as "Level 0". Mild attenuation: The amplitude decreases by 5% - 20% (e.g., reference 100 mV, defect 85 mV), marked as "Level 1", which may correspond to small-sized pores. Moderate attenuation: The amplitude decreases by 20% - 35% (e.g., reference 100 mV, defect 65 mV), marked as "Level 2", usually for lack of fusion or medium cracks. Severe attenuation: The amplitude decreases by more than 35% (e.g., reference 100 mV, defect below 60 mV), marked as "Level 3", indicating large-area defects or through cracks.

[0103] If the signal amplitude at the defect position is abnormally higher than that at the reference position (after excluding equipment failures), it may be due to enhanced interface reflection (such as the acoustic impedance difference between the base metal and the weld metal). It is necessary to further judge whether it is a pseudo-defect by combining time-frequency domain characteristics (such as phase shift). For multiple adjacent detection points of the same defect (such as 3 points within a 5 mm×5 mm grid), the average value of the attenuation coefficient is taken as the final attenuation index of the defect to avoid single-point noise interference.

[0104] In the embodiments of the present invention, by comparing the stress wave intensities at the defect position and the reference position, the attenuation coefficient directly reflects the hindering effect of the defect on the sound wave propagation: Distinguishing pores / cracks: Dense pores result in a relatively high attenuation coefficient (such as > 30%) due to the scattering effect, while cracks cause an abnormal attenuation coefficient (accompanied by phase shift) due to interface reflection, which can effectively distinguish the hazard levels of different defect types; the attenuation coefficient is positively correlated with the defect volume / area (e.g., 15% attenuation for pores of 5 mm 3 and 35% attenuation for a lack of fusion surface of 10 mm), providing a quantitative basis for defect hazard classification (traditional methods only qualitatively judge "defects exist").

[0105] The stress wave attenuation is directly related to the material continuity, and the attenuation coefficient can characterize the node stiffness change in real time: For every 10% increase in the attenuation coefficient, the corresponding node stiffness decreases by about 5% (verified by finite element simulation). For example, when the attenuation coefficient of a certain node exceeds the threshold (25%), the system automatically prompts insufficient stiffness, replacing the time-consuming detection of traditional loading tests (shortened from 2 weeks to 1 minute); by combining historical attenuation data, tracking the progressive impact of defect development on the structural performance (such as a 5% annual increase in the attenuation coefficient of a certain crack, indicating an exacerbation of interface separation), providing real-time data support for operation and maintenance decisions.

[0106] As a bridge between the physical properties of defects and the mechanical properties of structures, the attenuation coefficient improves the accuracy of life prediction: By studying the correlation between the attenuation coefficient and the structural failure time in historical data (e.g., the service life of nodes with attenuation > 30% is shortened by 25% compared to normal nodes), the limitations of traditional empirical formulas are broken through, and the prediction error is controlled within ±10%. Combining parameters such as load and humidity, the attenuation coefficient can quantify the deterioration rate of defects under complex working conditions (e.g., when the humidity is 80%, the growth rate of the attenuation coefficient increases by 20%), realizing personalized life prediction. The attenuation coefficient is directly calculated based on the ultrasonic echo signal without the need for additional sensors or offline testing. The detection time for a single node is maintained within 5 minutes, which is suitable for batch component detection of prefabricated buildings. Through signal calibration at the reference position (defect-free area), the influence of environmental noise is suppressed (effective calculation can still be maintained when the signal-to-noise ratio ≥ 15 dB), and the misjudgment rate is reduced by 40% compared to traditional methods.

[0107] In a preferred embodiment of the present invention, the dynamic offset parameters are input into a pre-trained welding quality assessment model to predict the structural life, including: Obtain the historical dynamic offset parameters of the cantilever beam - balcony slab welding joints in different prefabricated buildings, specifically including: collect the detection data of the cantilever beam - balcony slab welding joints from more than 50 past prefabricated building projects, including components that have been in use for 10 - 30 years (covering different regions, environmental humidity / temperature, and load conditions). Three types of dynamic parameters are extracted for each node: Defect depth gradient: The change rate of the defect depth along the length direction of the cantilever beam (e.g., the depth of a certain crack increases by 10 mm in the range of 1500 - 1510 mm on the X-axis, and the gradient is 1 mm / mm); Lateral offset vector: The planar distance and direction of the defect center from the theoretical axis (e.g., 12 mm in the positive direction of the Y-axis); Stress wave attenuation coefficient: The signal intensity difference between the defect position and the reference position (e.g., an attenuation of 30% indicates a significant loss of stress wave energy).

[0108] More than 2000 effective data are collected cumulatively, including parameter combinations of different defect types such as pores, lack of fusion, and cracks. Outliers (such as obvious equipment error points with a depth gradient > 5 mm / mm) and duplicate detection data (take the average value for multiple detections at the same position of the same node) are removed; they are classified and stored according to projects, component types (such as precast concrete balcony slabs / steel structure cantilever beams), and defect types to form a structured data set (Excel / CSV format).

[0109] Through accelerated fatigue tests, obtain the actual service life data of the welding joints as labels, specifically including: replicate the actual welding joints at a ratio of 1:1, use the same base material (Q345B steel) and welding process (such as gas shielded welding, current 200 A, voltage 25 V) to ensure consistency with the historical data samples. Loading scheme: Fatigue load: simulate the balcony slab live load (2.5kN / m²) + wind load combination, using sinusoidal loading (frequency 5Hz, stress ratio R=0.1), the load amplitude is 70% of the design bearing capacity; Environmental coupling: Some specimens were placed in a hot and humid chamber (temperature 40°C, humidity 90% RH) to simulate the corrosion environment in coastal areas, and some were placed in a room temperature dry environment as a control.

[0110] When the specimen has visible cracks (length > 5 mm) or the stiffness drops by more than 20% (the displacement sensor monitors the sudden change in the deflection of the cantilever end), it is judged as failure and the cumulative number of loading cycles is recorded (converted to the actual service life, such as 10 6 Cycle ≈ 15 years). Each specimen is tested three times in parallel, and the average life is taken as the label (for example, the three lives of a cracked specimen are 22 years, 24 years, and 23 years, and the label is 23 years); for retired nodes in historical projects (such as components removed after 25 years of use), the remaining life is inferred through on-site load tests to supplement the actual service life data (avoiding the extrapolation error of relying solely on accelerated tests).

[0111] The historical dynamic migration parameters are scaled to the interval [0, 1], the life data is logarithmically transformed, and the mean square error is used to measure the difference between the life data and the true life. Specifically, the depth gradient (usually 0-5mm / mm), lateral migration vector (0-20mm), and attenuation coefficient (0-50%) are uniformly scaled to the interval [0, 1]: Depth gradient = (actual value ÷ 5mm / mm) → e.g. 1mm / mm → 0.2, 5mm / mm → 1; Lateral offset = (actual distance ÷ 20mm) → e.g. 12mm → 0.6, 20mm → 1 (the negative direction is marked separately with a sign, such as -10mm → -0.5 on the Y axis, normalized after absolute value processing); Attenuation coefficient = (actual value ÷ 50%) → e.g. 30% → 0.6, 50% → 1.

[0112] Use the Min-Max normalization method to retain the relative differences between parameters (avoid the Z-score standardization to distort the distribution of small samples). Perform a logarithmic transformation on the life labels (10-50 years) (e.g. ln(20 years) = 3, ln(50 years) = 3.91) to transform the right-skewed distribution into an approximate normal distribution and improve the neural network fitting effect; divide the training set and the validation set (8:2 ratio) to ensure that the samples of different defect types and life intervals are balanced (e.g. samples with life <20 years and ≥20 years each account for 50%). The fully connected neural network is trained to obtain a trained fully connected neural network, specifically including: network architecture design: Input layer: 3 neurons (corresponding to 3 dynamic parameters: depth gradient, lateral offset, attenuation coefficient); Hidden layer: 2 layers (64 neurons per layer), using ReLU activation function (to capture the non-linear relationship between parameters), adding a Dropout layer (inactivation rate 0.2) between layers to prevent overfitting; Output layer: 1 neuron (outputting the logarithm-transformed life prediction value), using a linear activation function (to adapt to the regression task). Training process: Optimizer, Adam (learning rate 0.001), dynamically adjusting the learning rate to balance the convergence speed and accuracy; Loss function: Mean Squared Error (MSE), calculating the mean of the squared differences between the predicted life and the actual life (e.g., if the predicted value ln(23) = 3.14 and the actual value ln(23) = 3.14, MSE = 0); Iteration strategy: Batch size 32, iterating 100 epochs; evaluating MSE on the validation set every 5 epochs. If the validation set error does not decrease for 10 consecutive epochs, early stopping is triggered to avoid overfitting (the actual training usually converges at 60 - 80 epochs). After training is completed, use the test set (200 data points independent of the training / validation set) to calculate the Mean Absolute Error (e.g., the average deviation between the predicted life and the actual life ≤ ±2 years), and plot the scatter plot of the predicted values and the actual values (ideally, the data points are distributed along the line y = x).

[0113] Based on the fully-connected neural network after training and the real-time dynamic offset parameters, to predict the structural life, specifically including: for the dynamic parameters obtained from real-time detection (such as the current depth gradient of a certain node is 0.8 mm / mm, lateral offset is 8 mm, attenuation coefficient is 25%), scale them according to the normalization method in the training stage (0.8 ÷ 5 = 0.16, 8 ÷ 20 = 0.4, 25% ÷ 50% = 0.5), forming the input vector [0.16, 0.4, 0.5]. Input the preprocessed parameters into the trained neural network, calculate the output value through forward propagation (such as obtaining the logarithmic life 3.25), and restore it to the actual life prediction value through exponential transformation (e 3.25 ≈ 25.7 years). The prediction result is accompanied by a confidence score (such as when the variance of the activation value of the output layer of the neural network < 0.05, it is marked as "high confidence"), docked with the BIM system, marking the remaining life of the node in the 3D model (such as red warning for < 15 years, yellow warning for 15 - 25 years, green display for > 25 years), and automatically generating maintenance suggestions (such as suggesting reinforcement within three months when the life < 10 years).

[0114] In the embodiments of the present invention, by using the historical dynamic offset parameters (such as depth gradient, lateral offset vector, attenuation coefficient) of welding joints in different projects, combined with the true life labels obtained from accelerated fatigue tests (such as the actual failure time of a certain joint is 25 years), a "defect parameter - life" mapping relationship is constructed, avoiding the rough estimation of traditional methods relying on material manuals or empirical formulas (such as the life estimation error of traditional methods often exceeds ±20%). The fully connected neural network automatically learns the complex non-linear relationship between dynamic parameters and life (such as when the stress wave attenuation coefficient increases by 10%, the life shortens by 15%; when the depth gradient > 1 mm / mm, the life impact weight increases by 30%). In a certain actual project, the life prediction error of the model for joints with dense pores is controlled within ±10%, and the accuracy is improved by 2 times compared with the traditional accelerated test extrapolation method (error ±30%).

[0115] By integrating the spatial position (lateral offset vector), extension trend (depth gradient), and mechanical properties (attenuation coefficient) of defects, the one-sidedness of single parameters is avoided. For example: Defects with only a lateral offset > 10 mm may be judged as "high risk", but combined with a depth gradient < 0.5 mm / mm (the defect does not extend to the core area) and an attenuation coefficient < 20% (the stiffness reduction is limited), the model will correct the life prediction (adjusted from 20 years to 25 years) to avoid over-reinforcement; Through training with historical data under different environments (temperature, humidity) and load conditions, the model can adjust the prediction logic for the service environment of specific joints (such as when the humidity > 80%, the impact weight of the attenuation coefficient on life increases by 20%), solving the defect that traditional methods cannot consider the environmental coupling effect. The single-joint life prediction takes less than 1 second (based on the forward propagation of the trained neural network), supporting the automated quality assessment of prefabricated components in prefabricated buildings (such as detecting 200 joints per hour on a production line and generating life reports synchronously), and the efficiency is improved by more than 1000 times compared with traditional test methods (it takes 72 hours to test a single joint).

[0116] As Figure 2 shown, the embodiments of the present invention also provide an intelligent quality inspection system for prefabricated buildings, including: A collection module for collecting ultrasonic echo signals of the welded joints between the cantilever beam end and the balcony slab; A fusion module for constructing a reference space grid model based on the theoretical installation coordinates of the embedded parts, generating a three-dimensional defect distribution map of the welding interface through the time-of-flight inversion algorithm of ultrasonic echo signals, and performing spatial registration with the reference grid model to obtain a defect-reference fusion model; A processing module, which is used to perform feature fusion analysis on ultrasonic echo signals by using a deep residual network based on the spatial coordinate mapping relationship of the defect-reference fusion model, so as to obtain a feature vector corresponding to each signal; according to the feature vector and the time-frequency domain joint threshold segmentation algorithm, identify the types of welding defects and mark the defect spatial coordinate set, forming a defect coordinate set containing four-dimensional information, and the welding defect types include pores, lack of fusion and cracks; A calculation module, which is used to calculate the dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model, and the parameters include defect depth gradient, lateral offset vector and stress wave attenuation coefficient; A prediction module, which is used to input the dynamic offset parameters into a pre-trained welding quality evaluation model to predict the structural life.

[0117] The above is the preferred embodiment of the present invention. It should be pointed out that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. An intelligent detection method for prefabricated building quality, characterized in that: The method comprises: Collect the ultrasonic echo signal of the welding node between the cantilever beam end and the balcony slab; A reference space grid model is constructed based on the theoretical installation coordinates of the embedded parts. A three-dimensional defect distribution map of the welding interface is generated by the time-of-flight inversion algorithm of the ultrasonic echo signal, and then spatially aligned with the reference grid model to obtain a defect-reference fusion model. Based on the spatial coordinate mapping relationship of the defect-reference fusion model, a deep residual network is used to perform feature fusion analysis on the ultrasonic echo signal to obtain the feature vector corresponding to each signal; according to the feature vector and the time-frequency domain joint threshold segmentation algorithm, the welding defect type is identified and the defect spatial coordinate set is marked to form a defect coordinate set containing four-dimensional information. The welding defect types include pores, lack of fusion and cracks; Calculating dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model, the parameters including defect depth gradient, lateral offset vector and stress wave attenuation coefficient; The dynamic offset parameters are input into a pre-trained weld quality assessment model to predict the structure life.

2. According to claim 1, a method for intelligent detection of prefabricated building quality is characterized in that: The reference space grid model is constructed based on the theoretical installation coordinates of the embedded parts, including: Extract the theoretical coordinates of embedded parts from the architectural design drawings, including the coordinates of each embedded part in three-dimensional space; determine the spatial range of the grid to be constructed based on the theoretical coordinates of the embedded parts; According to the resolution of ultrasonic testing and the spatial range of the grid, the size of the grid unit is set, and a three-dimensional rectangular coordinate system is established in the detection area with the theoretical coordinates of the embedded parts as reference, where the origin is the geometric center of the embedded parts, the X-axis is along the length direction of the cantilever beam, the Y-axis is along the width direction of the balcony plate, and the Z-axis is along the thickness direction of the welding interface; Starting from the origin, the grid lines are evenly divided along the X-axis, Y-axis and Z-axis to form a regular cubic grid; The theoretical geometric features of the welding nodes are embedded in the cubic grid to form a reference space grid model containing the coordinates of all grid points and the theoretical geometric features, including the theoretical position of the welding interface, the theoretical axis of the cantilever beam, and the theoretical contour of the embedded parts.

3. The intelligent detection method for prefabricated building quality according to claim 2 is characterized in that: The three-dimensional defect distribution map of the welding interface is generated by the time-of-flight inversion algorithm of the ultrasonic echo signal, and is spatially registered with the reference grid model to obtain the defect-reference fusion model, including: The probe is attached to the surface of the cantilever beam, emits high-frequency ultrasonic pulses to the welding interface, and receives the reflected echo signal. Each scan records the echo signal of a position, including the signal amplitude and flight time; According to the propagation speed of ultrasonic waves in the material, the flight time of the echo signal is converted into the distance between the defect and the probe; According to the actual position and beam angle during scanning, the coordinates of the defect in three-dimensional space are calculated. The defect coordinates obtained from each scan are used as a data point. After multiple scans, point cloud data containing the defect points is formed. Map the point cloud data to a three-dimensional space grid, divide the cube voxels according to the preset accuracy, and the gray value of each voxel represents the defect probability of the corresponding area, finally forming a three-dimensional defect distribution map; Extract the four corner points of the embedded parts and the edge line of the welding interface from the reference space grid model and the three-dimensional defect distribution map; Match the coordinates of the corner points of the embedded parts with the coordinates of the theoretical corner points in the reference grid, calculate the preliminary translation and rotation angle to align the overall positions of the two, calculate the distance from each measured defect point to the nearest theoretical point in the reference grid as the error, and gradually adjust the position and posture of the three-dimensional defect distribution map until the overall error is less than the set threshold to form a registered defect map; The registered defect map is embedded into the benchmark grid model so that the coordinates of each defect point correspond to the spatial position of the benchmark grid. Finally, a defect-benchmark fusion model containing theoretical geometric features and measured defect data is obtained.

4. The intelligent detection method for prefabricated building quality according to claim 3 is characterized in that: Based on the spatial coordinate mapping relationship of the defect-reference fusion model, a deep residual network is used to perform feature fusion analysis on the ultrasonic echo signal to obtain the feature vector corresponding to each signal, including: Obtain the original ultrasonic echo signal, each signal corresponds to a detection position, and includes a curve of amplitude variation over time; Perform frequency domain analysis on ultrasonic echo signals to generate time-frequency diagrams and extract energy distribution characteristics of signals in different frequency bands. In the time-frequency diagram, the horizontal axis is time and the vertical axis is frequency. The detection position coordinates corresponding to each signal, integrating the ultrasonic echo signal, time-frequency diagram and position coordinates into an input tensor; Based on the input tensor and the trained deep residual network, a comprehensive feature vector containing time-frequency domain signal features and spatial position information is generated.

5. The intelligent detection method for prefabricated building quality according to claim 4 is characterized in that: According to the feature vector and the time-frequency domain joint threshold segmentation algorithm, the welding defect type is identified and the defect space coordinate set is marked to form a defect coordinate set containing four-dimensional information. The welding defect type includes pores, lack of fusion and cracks, including: From the feature vector, the key parameters related to the time-frequency domain characteristics of the signal are separated, including time-domain characteristics and frequency-domain characteristics. The time-domain characteristics include peak amplitude, rising edge time and duration; the frequency-domain characteristics include center frequency, bandwidth, energy center frequency and characteristic frequency components. The time domain features and frequency domain features of each detection signal are integrated into a matrix. Each row of the matrix corresponds to a signal feature vector at a detection position, including the time domain and frequency domain parameters of the signal. Each column of the matrix corresponds to a specific feature parameter, forming a multi-dimensional feature space. Based on historical inspection data and defect sample training, the joint judgment conditions of time domain features and frequency domain features are determined for three types of defects: pores, lack of fusion and cracks. According to the characteristic parameters of each detection signal and the joint judgment conditions, the characteristic vectors of each signal in the matrix are judged one by one to obtain the defect signal and the detection position coordinates corresponding to the defect signal; According to the spatial coordinate mapping relationship of the defect-benchmark fusion model, the detection position coordinates corresponding to the defect signal are converted into three-dimensional coordinates in the defect-benchmark fusion model, and the corresponding welding defect type label is added to each defect coordinate to form a defect coordinate set containing four-dimensional information.

6. The intelligent detection method for prefabricated building quality according to claim 5 is characterized in that: The dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam are calculated in the defect-reference fusion model, and the parameters include the defect depth gradient, the lateral offset vector and the stress wave attenuation coefficient, including: Extract the information of the theoretical axis of the cantilever beam from the defect-reference fusion model, including the direction, position and related geometric parameters of the axis; determine the defect coordinate set, each defect coordinate contains its position information in three-dimensional space; According to the coordinate system and geometric features in the defect-reference fusion model, the depth direction is determined, and the depth direction is along the direction perpendicular to the cantilever beam surface and pointing to the inside of the weld; The defect coordinate sets are grouped according to their positions in the length direction of the cantilever beam, and for each defect in each group, its coordinate value in the depth direction is calculated; For defects in adjacent groups, the ratio of their coordinate difference in the depth direction to the distance difference in the cantilever beam length direction is calculated, that is, the defect depth gradient.

7. The intelligent detection method for prefabricated building quality according to claim 6 is characterized in that: The process of determining the lateral offset vector is: According to the coordinate system in the defect-reference fusion model and the geometric information of the cantilever beam, the transverse direction is determined, and the transverse direction is a direction perpendicular to the theoretical axis of the cantilever beam and in the plane where the cantilever beam is located; For each defect point in the defect coordinate set, project it onto the theoretical axis of the cantilever beam to obtain the corresponding projection point; The vector between each defect point and its projection point is calculated, that is, the lateral offset vector of the defect point relative to the theoretical axis of the cantilever beam, which includes the size and direction information of the offset.

8. The intelligent detection method for prefabricated building quality according to claim 7 is characterized in that: The process of determining the stress wave attenuation coefficient is: Using the principle of ultrasonic testing and related physical models, the propagation process of stress waves in the cantilever beam and welding node is simulated, and the intensity of stress waves is measured at each defect position in the defect coordinate set and the reference position on the cantilever beam; The stress wave attenuation coefficient is calculated based on the change in the intensity of the stress wave during its propagation.

9. The intelligent detection method for prefabricated building quality according to claim 8 is characterized in that: Dynamic offset parameters are fed into a pre-trained weld quality assessment model to predict structure life, including: Obtain the historical dynamic offset parameters of cantilever beam-balcony slab welding nodes in different prefabricated buildings; Through accelerated fatigue testing, the actual service life data of welding nodes is obtained as labels; The historical dynamic offset parameters are scaled to the interval [0, 1], the life data are logarithmically transformed, and the mean square error is used to measure the difference between the life data and the true life; Training the fully connected neural network to obtain a trained fully connected neural network; The structural life is predicted based on the trained fully connected neural network and real-time dynamic offset parameters.

10. An intelligent detection system for prefabricated building quality, characterized in that: The system implements the method according to any one of claims 1 to 9, comprising: An acquisition module is used to collect ultrasonic echo signals of the welding nodes between the cantilever beam end and the balcony slab; The fusion module is used to construct a reference space grid model based on the theoretical installation coordinates of the embedded parts, generate a three-dimensional defect distribution map of the welding interface through the time-of-flight inversion algorithm of the ultrasonic echo signal, and perform spatial registration with the reference grid model to obtain a defect-reference fusion model; A processing module is used to perform feature fusion analysis on ultrasonic echo signals using a deep residual network based on the spatial coordinate mapping relationship of the defect-reference fusion model to obtain a feature vector corresponding to each signal; according to the feature vector and the time-frequency domain joint threshold segmentation algorithm, the welding defect type is identified and the defect spatial coordinate set is marked to form a defect coordinate set containing four-dimensional information, wherein the welding defect type includes pores, lack of fusion and cracks; A calculation module, used to calculate the dynamic offset parameters of the defect coordinate set relative to the theoretical axis of the cantilever beam in the defect-reference fusion model, wherein the parameters include the defect depth gradient, the lateral offset vector and the stress wave attenuation coefficient; A prediction module is used to input the dynamic offset parameters into a pre-trained weld quality assessment model to predict the structure life.

Images