Crack detection device for constructional engineering and use method

Abstract

The invention belongs to the technical field of building health detection, and particularly relates to a crack detection device for building engineering and a use method, and the crack detection device comprises a data acquisition module which is configured to obtain multi-modal nondestructive detection information of a building structure; the data processing module is configured to perform matching fusion on different modal data in the standardized multi-modal detection data set in time and space, and perform association and integration on feature information of the different modal data in the standardized multi-modal detection data set to obtain a cross-modal feature fusion tensor; the crack recognition and classification module is configured to recognize and classify cracks based on an improved deep learning model according to the input cross-modal feature fusion tensor and output a crack detection result; and the intelligent early warning module is configured to perform quantitative analysis on the crack detection result according to the output crack detection result and send out early warning information according to a preset threshold value.

Description

Technical Field

[0001] The present invention belongs to the technical field of building health detection, and in particular relates to a crack detection device for construction engineering and a method of use. Background Art

[0002] In the field of construction engineering, crack detection in concrete structures (such as bridges, tunnels, and high-rise buildings) is a core component of ensuring structural safety. Traditional detection methods rely primarily on manual visual observation combined with crack width gauges. This method has significant limitations: on the one hand, manual inspection efficiency is extremely low and is significantly affected by the experience and subjective judgment of the inspector; on the other hand, manual inspection can only identify visible cracks on the surface and cannot detect internal defects in the concrete (such as honeycomb damage caused by steel corrosion and hollowing of the grouting layer), which are often potential major hazards to structural safety. To this end, an improved crack detection device and method for use in construction engineering have been designed. Summary of the Invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a crack detection device for construction engineering and a method of use thereof to solve the problems in the above-mentioned background technology.

[0004] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0005] A crack detection device for construction engineering, comprising: a data acquisition module, wherein the data acquisition module is configured to obtain multimodal nondestructive testing information of a building structure;

[0006] a data preprocessing module configured to preprocess the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set;

[0007] a data processing module configured to match and fuse different modal data in the standardized multimodal detection dataset in time and space, and to associate and integrate feature information of different modal data in the standardized multimodal detection dataset to obtain a cross-modal feature fusion tensor;

[0008] The crack identification and classification module is configured to identify and classify cracks based on the input cross-modal feature fusion tensor based on the improved deep learning model and output the crack detection results;

[0009] The intelligent early warning module is configured to perform quantitative analysis on the crack detection results according to the output crack detection results, and issue early warning information according to a preset threshold value.

[0010] Furthermore, the data acquisition module includes a visual image acquisition module for acquiring image information of the building structure surface at the same moment, an ultrasonic signal acquisition module for acquiring internal structure information of the building structure at the same moment, an infrared data acquisition module for acquiring temperature information of the building structure surface at the same moment, and a synchronous trigger module. The multimodal nondestructive testing information includes visual images, ultrasonic signals and temperature information of the building structure surface.

[0011] Furthermore, the synchronization trigger module includes a hardware timer, the visual image acquisition module includes an industrial camera, the ultrasonic signal acquisition module includes a phased array ultrasonic probe, the infrared data acquisition module includes an infrared thermal imager, and the hardware timer is electrically connected to the industrial camera, the phased array ultrasonic probe and the infrared thermal imager.

[0012] Furthermore, the preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes:

[0013] Visual image preprocessing first uses bilateral filtering to denoise the image, then performs morphological opening operation to remove tiny noise points, then uses Frangi vascular filtering algorithm to enhance crack features, and finally uses adaptive threshold segmentation to extract suspected crack area ROI.

[0014] Furthermore, the preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes:

[0015] Ultrasonic signal processing includes three steps: time domain filtering, phase correction, and feature parameter extraction. Specifically, time domain filtering is to perform preliminary purification of the ultrasonic signal, and a 50-5000Hz bandpass filter is applied to remove environmental noise; phase correction is to eliminate the phase offset caused by the coupling difference between the probe and the surface through Hilbert transform de-envelope; feature parameter extraction is to extract the arrival time t of the defect echo from the ultrasonic signal. p , peak intensity A p , rise time t r， Construct ultrasound feature vector [t p ,A p ,t r ].

[0016] Furthermore, the preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes:

[0017] Infrared data processing includes two steps: temperature normalization and Gaussian pyramid downsampling. Specifically, the collected temperature value is converted into the difference relative to the ambient temperature (ΔT = TT env), improve the visual recognition of temperature distribution through pseudo-color mapping; use Gaussian blur or median filtering to smooth noise, segment abnormal hot areas according to temperature thresholds, and locate the detection area in combination with the building structure model.

[0018] Furthermore, the data processing module includes a spatiotemporal registration module configured to obtain the external parameters of the visual camera, including the rotation matrix R and the translation vector T, and convert the pixel coordinates in the visual image into points in the camera coordinate system; obtain the position information of the ultrasound probe, and calculate the position information of the ultrasound probe by the formula

[0019] P world =R·P camera +T=R′·P ultrasonic +T′ unifies the visual and ultrasound data into the world coordinate system, where P camera is the camera coordinate system point converted from visual pixel coordinates, P ultrasonic is the ultrasound scanning coordinate.

[0020] Furthermore, the data processing module includes a feature extraction and association module, which is configured to extract edge, texture, and color features from the visual image; extract the arrival time, peak intensity, and rise time features of the defect echo from the ultrasonic signal; extract the temperature difference ΔT and thermal diffusivity feature α from the infrared data; and for each visual pixel point (x, y) belonging to the crack ROI, associate it with the ultrasonic feature (t p ,A p ) and infrared features (ΔT, α) are associated to construct a three-dimensional feature vector [g(x, y), u(x, y), i(x, y)], and a cross-modal feature fusion tensor is obtained, where g(x, y) is the visual feature, u(x, y) is the ultrasonic feature, and i(x, y) is the infrared feature. In order to highlight the key modal features, the attention mechanism is used to assign feature weights: w = σ(W g g+W u u+W i i), where σ is the Sigmoid function, W g 、W u 、W i are the weight matrices of visual, ultrasonic and infrared features respectively.

[0021] Furthermore, the crack recognition and classification module is configured to identify and classify cracks based on the input cross-modal feature fusion tensor based on the improved deep learning model, and output the crack detection results, including encoding the fused feature vector [g,u,i] into a pseudo-color image (R channel = visual probability, G channel = ultrasonic echo intensity, B channel = infrared ΔT) as the input of the YOLOv8 model. At the same time, an auxiliary input branch is added to directly input the ultrasonic C-scan grayscale image and infrared thermal image to form a multi-branch network structure; a cross-modal feature interaction module is introduced in the neck network (PAFPN): Among them F vis is the visual feature, F ult For ultrasonic features, the defect correlation in visual features is enhanced by channel-level attention, and the YOLOv8 model outputs crack type, confidence, and geometric parameters for each detection box.

[0022] Furthermore, the crack detection results are quantitatively analyzed based on the output, and warning information is issued according to a preset threshold, including: measurement of crack length and width and calculation of curvature, calculation of internal defect area of ​​the crack and analysis of thermal diffusivity.

[0023] A method for using a crack detection device for construction engineering includes the following steps: a first step of obtaining multimodal nondestructive testing information of a building structure;

[0024] The second step is to preprocess the multimodal nondestructive testing information obtained by the data acquisition module to obtain a standardized multimodal testing data set;

[0025] Step 3: Based on the input cross-modal feature fusion tensor, cracks are identified and classified based on the improved deep learning model, and the crack detection results are output;

[0026] Step 4: Based on the output crack detection results, perform quantitative analysis on the crack detection results and issue early warning information based on the preset threshold.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] 1. By integrating multimodal data of visual images, ultrasonic signals, and infrared thermal images, it realizes the integrated detection of surface cracks, internal defects, and thermal bridge effects, solving the problem that traditional manual detection can only identify surface cracks.

[0029] 2. This invention breaks through the efficiency and accuracy bottlenecks of traditional detection technologies through the spatiotemporal coordination of multimodal data and cross-modal optimization of deep learning models, providing an "efficient, accurate, and intelligent" integrated solution for crack detection in construction projects, which has significant engineering application value and industry promotion significance. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 This is a diagram showing the composition of a crack detection device for construction engineering according to the present invention;

[0031] Figure 2 This is a structural diagram of a data acquisition module in a crack detection device for construction engineering according to the present invention;

[0032] Figure 3 This is another diagram showing another component of a data acquisition module in a crack detection device for construction engineering according to the present invention;

[0033] Figure 4 This is a diagram showing how to use a crack detection device for construction engineering according to the present invention. DETAILED DESCRIPTION

[0034] In order to enable those skilled in the art to better understand the present invention, the technical solution of the present invention is further described below with reference to the accompanying drawings and embodiments.

[0035] Among them, the drawings are only used for illustrative purposes and represent only schematic diagrams rather than actual pictures, and should not be understood as limiting this patent; in order to better illustrate the embodiments of the present invention, some parts of the drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product; for those skilled in the art, it is understandable that some well-known structures and their descriptions in the drawings may be omitted.

[0036] Example 1:

[0037] like Figure 1-3 As shown, the present invention. Specifically, a crack detection device for construction engineering includes a data acquisition module, wherein the data acquisition module is configured to obtain multimodal nondestructive testing information of the building structure;

[0038] a data preprocessing module configured to preprocess the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set;

[0039] a data processing module configured to match and fuse different modal data in the standardized multimodal detection dataset in time and space, and to associate and integrate feature information of different modal data in the standardized multimodal detection dataset to obtain a cross-modal feature fusion tensor;

[0040] The crack identification and classification module is configured to identify and classify cracks based on the input cross-modal feature fusion tensor based on the improved deep learning model and output the crack detection results;

[0041] The intelligent early warning module is configured to perform quantitative analysis on the crack detection results according to the output crack detection results, and issue early warning information according to a preset threshold value.

[0042] Furthermore, the data acquisition module includes a visual image acquisition module for acquiring image information of the building structure surface at the same moment, an ultrasonic signal acquisition module for acquiring internal structure information of the building structure at the same moment, an infrared data acquisition module for acquiring temperature information of the building structure surface at the same moment, and a synchronous trigger module. The multimodal nondestructive testing information includes visual images, ultrasonic signals and temperature information of the building structure surface.

[0043] Furthermore, the synchronization trigger module includes a hardware timer, the visual image acquisition module includes an industrial camera, the ultrasonic signal acquisition module includes a phased array ultrasonic probe, the infrared data acquisition module includes an infrared thermal imager, and the hardware timer is electrically connected to the industrial camera, the phased array ultrasonic probe and the infrared thermal imager.

[0044] To ensure temporal and spatial consistency of data collected by different devices, a hardware timer (such as an FPGA generating synchronization pulses) is used to achieve microsecond-level synchronization of industrial cameras, phased array ultrasonic probes, and infrared thermal imagers. This high-precision synchronization mechanism can avoid data misalignment caused by minor structural deformations (such as thermal expansion and contraction caused by temperature). If the images captured by the camera are not synchronized in time with the data collected by the ultrasonic probe and infrared thermal imager, then in subsequent analysis, normal structural changes may be mistaken for cracks, or some real crack information may be missed. Therefore, the synchronous trigger mechanism is like setting a unified "beat" for the entire detection system, enabling various devices to work together and provide accurate and reliable raw data for subsequent multimodal fusion.

[0045] Furthermore, the preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes:

[0046] Visual image preprocessing first uses bilateral filtering to denoise the image, then performs morphological opening operation to remove tiny noise points, then uses Frangi vascular filtering algorithm to enhance crack features, and finally uses adaptive threshold segmentation to extract suspected crack area ROI.

[0047] Visual image preprocessing is a complex and sophisticated process, consisting of multiple steps, each with a specific purpose. First, bilateral filtering is used to denoise the image. Bilateral filtering is a nonlinear filtering method that effectively preserves edge information while removing Gaussian noise. By setting appropriate parameters (such as d = 5, sigmaColor = 50, and sigmaSpace = 50), image noise can be effectively removed while maintaining the clarity of edge features such as cracks. Next, a morphological opening operation is performed to remove small noise points. This morphological opening operation, which first performs an erosion operation followed by a dilation operation, can remove small particles of noise while preserving the shape of larger objects. Using an elliptical structuring element better adapts to the shape of cracks, removing small noise points unrelated to the cracks and enhancing image clarity. Finally, the Frangi filter algorithm is used to enhance crack features. The Frangi filter is an algorithm specifically designed for detecting linear structures in images. It enhances cracks based on their geometric characteristics (such as length, width, and curvature). By setting different scaling parameters, cracks of varying widths can be detected, thereby improving crack recognition. Finally, adaptive threshold segmentation is used to extract the suspected crack region (ROI). This method automatically determines the threshold based on the local features of the image, dividing the image into two parts: the target (crack) and the background. The Otsu algorithm automatically calculates the optimal threshold, making the segmentation more accurate. The extracted ROI includes areas where cracks are likely to exist, providing a focus for subsequent analysis.

[0048] Furthermore, the preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes:

[0049] Ultrasonic signal processing includes three steps: time domain filtering, phase correction, and feature parameter extraction. Specifically, time domain filtering is to perform preliminary purification of the ultrasonic signal, and a 50-5000Hz bandpass filter is applied to remove environmental noise; phase correction is to eliminate the phase offset caused by the coupling difference between the probe and the surface through Hilbert transform de-envelope; feature parameter extraction is to extract the arrival time t of the defect echo from the ultrasonic signal. p , peak intensity A p , rise time t r , construct the ultrasound feature vector [t p ,A p ,t r ].

[0050] Ultrasonic signal processing primarily involves three steps: time-domain filtering, phase correction, and feature parameter extraction. Time-domain filtering provides preliminary purification of the ultrasonic signal by applying a 50-5000Hz bandpass filter (Butterworth IIR) to remove ambient noise (e.g., mechanical vibration frequencies <50Hz). By setting appropriate filtering parameters, low-frequency mechanical vibration noise and high-frequency interference signals can be effectively removed, improving ultrasonic signal quality. Phase correction uses Hilbert transform de-envelopment to eliminate phase shifts caused by differential coupling between the probe and the surface. In ultrasonic testing, the coupling between the probe and the building surface affects the propagation and reflection of ultrasonic waves, resulting in changes in the signal phase. The Hilbert transform converts the ultrasonic signal from the time domain to the frequency domain. De-envelopment then extracts the signal envelope information, eliminating the effects of phase shifts and ensuring that the signal more accurately reflects the internal structural conditions of the building. Feature parameter extraction extracts characteristic information related to cracks from the ultrasonic signal, such as the arrival time (tp), peak intensity (Ap), and rise time (tr) of the defect echo. These characteristic parameters can reflect information such as the location, size, and properties of the crack. By constructing the ultrasonic characteristic vector [tp, Ap, tr], these characteristic information can be integrated to provide a basis for subsequent analysis and judgment.

[0051] Furthermore, the preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes:

[0052] Infrared data processing includes two steps: temperature normalization and Gaussian pyramid downsampling. Specifically, the collected temperature value is converted into the difference relative to the ambient temperature (ΔT = TT env ), improve the visual recognition of temperature distribution through pseudo-color mapping; use Gaussian blur or median filtering to smooth noise, segment abnormal hot areas according to temperature thresholds, and locate the detection area in combination with the building structure model.

[0053] Infrared data processing mainly includes two steps: temperature normalization and Gaussian pyramid downsampling. Temperature normalization is to convert the collected temperature value into the difference relative to the ambient temperature (ΔT = TT env) to eliminate the influence of ambient temperature drift. In actual detection, changes in ambient temperature will interfere with the detection results of the infrared thermal imager. By calculating the relative temperature difference, the influence of ambient temperature can be eliminated, so that the detection results can more accurately reflect the temperature anomaly on the building surface. For example, when the temperature change is ≤1°C during the detection period, the ΔT error is ≤0.2°C, which can ensure the reliability of the temperature data. Gaussian pyramid downsampling is an image downsampling method. Using Gaussian pyramid downsampling (scale factor 2) can reduce the amount of calculation while retaining thermal anomaly areas ≥2mm in size. When processing large amounts of infrared thermal imaging data, the amount of calculation is an important issue. Downsampling can reduce the amount of data and improve processing efficiency. At the same time, retaining a certain size of thermal anomaly area can ensure that important information related to cracks is not missed.

[0054] The ROI extracted through visual preprocessing provides a targeted analysis area for the ultrasonic and infrared data, a key indicator of the inter-step linkage. In actual inspections, building surfaces are typically large, and comprehensive analysis of both ultrasonic and infrared data would consume significant computing resources and time. However, the ROI extracted through visual preprocessing clearly identifies potential crack locations, allowing for in-depth analysis of only the ultrasonic A-scan signals and infrared sub-areas corresponding to the ROI. This reduces unnecessary computation and improves inspection efficiency. This creates a cascaded processing logic of "visual coarse screening → ultrasonic / infrared fine inspection," making the entire inspection process more efficient and accurate.

[0055] Furthermore, the data processing module includes a spatiotemporal registration module configured to obtain the extrinsic parameters of the visual camera, including the rotation matrix R and the translation vector T, and convert the pixel coordinates in the visual image into points in the camera coordinate system; obtain the position information of the ultrasound probe through the formula;

[0056] Pworld=R·Pcamera+T=R′·Pultrasonic+T′ unifies the visual and ultrasonic data into the world coordinate system, where Pcamera is the camera coordinate system point converted from the visual pixel coordinate, and Pultrasonic is the ultrasonic scanning coordinate.

[0057] Furthermore, the data processing module includes a feature extraction and association module, which is configured to extract edge, texture, and color features from the visual image; extract the arrival time, peak intensity, and rise time features of the defect echo from the ultrasonic signal; extract the temperature difference ΔT and thermal diffusivity feature α from the infrared data; and for each visual pixel point (x, y) belonging to the crack ROI, associate it with the ultrasonic feature (t p ,A p) and infrared features (ΔT, α) are associated to construct a three-dimensional feature vector [g(x, y), u(x, y), i(x, y)], and a cross-modal feature fusion tensor is obtained, where g(x, y) is the visual feature, u(x, y) is the ultrasonic feature, and i(x, y) is the infrared feature. In order to highlight the key modal features, the attention mechanism is used to assign feature weights: w = σ(W g g+W u u+W i i), where σ is the Sigmoid function, W g 、W u 、W i are the weight matrices of visual, ultrasonic and infrared features respectively.

[0058] By unifying data from different modalities into the same coordinate system through spatiotemporal registration, spatial consistency is provided for feature cascading. Feature cascading enables the local features of a single modality (such as edges in vision) to complement the global features of other modalities (such as depth in ultrasound), thereby improving the accuracy of crack classification. For example, in a visual image, some surface stains or textures may be misjudged as cracks, but by combining the depth information in the ultrasound features, it is possible to determine whether the area actually has internal defects, thereby eliminating misjudgments. This close connection between steps enables data from different modalities to complement and verify each other, providing a more reliable basis for subsequent crack classification and quantitative evaluation.

[0059] Furthermore, the crack recognition and classification module is configured to identify and classify cracks based on the input cross-modal feature fusion tensor based on the improved deep learning model, and output the crack detection results, including encoding the fused feature vector [g,u,i] into a pseudo-color image (R channel = visual probability, G channel = ultrasonic echo intensity, B channel = infrared ΔT) as the input of the YOLOv8 model. At the same time, an auxiliary input branch is added to directly input the ultrasonic C-scan grayscale image and infrared thermal image to form a multi-branch network structure; a cross-modal feature interaction module is introduced in the neck network (PAFPN): Among them F vis is the visual feature, F ult For ultrasonic features, the defect correlation in visual features is enhanced by channel-level attention, and the YOLOv8 model outputs crack type, confidence, and geometric parameters for each detection box.

[0060] The multi-branch network structure allows the model to analyze data from different perspectives, improving its generalization and accuracy. Ultrasonic C-scans provide detailed structural information about the building's interior, while infrared thermal images reveal surface temperature anomalies. This information complements the pseudo-color image, enabling the model to gain a more comprehensive understanding of the building's structural condition. Channel-level attention is used to enhance defect correlation in visual features. The cross-modal feature interaction module facilitates information exchange and fusion between features from different modalities. In traditional models, features from different modalities are often processed independently, lacking effective interaction. This module, however, allows ultrasonic features to adjust and enhance visual features, focusing more on information related to cracks. Ultrasonic features can indicate the presence of internal defects in certain areas, allowing the visual features to more carefully analyze these areas and improve crack recognition. The model outputs the crack type (surface, internal, other), confidence score (0-1), and geometric parameters (length L, width W, depth D) for each detection box. Determining the crack type helps determine its severity and treatment. Surface cracks may require simple repairs, while internal cracks may require more in-depth inspection and treatment. The confidence level indicates the model's confidence in the test results, and users can use this confidence level to determine the reliability of the test results. Geometric parameters (length L, width W, depth D) can quantify the size and location of cracks, providing specific data support for subsequent repairs and reinforcements.

[0061] Depth D is valid for internal defects and is calculated using the formula: D = v·tp / 2, where the velocity of sound v is dynamically adjusted based on the concrete's rebound value (e.g., a rebound value of 35 MPa corresponds to v = 3900 m / s). The velocity of sound in concrete is affected by factors such as its strength. By dynamically adjusting the velocity of sound based on the rebound value, the depth of internal defects can be more accurately calculated. Using a fixed velocity of sound for calculations may result in inaccurate depth calculations, thus affecting the assessment and treatment of internal defects.

[0062] The preprocessed fused features serve as model input, avoiding interference from single-modal noise. In actual inspections, single-modal data may be affected by various noise factors, resulting in inaccurate inspection results. However, through the fusion and preprocessing of multimodal data, noise can be effectively removed and the quality of the input data can be improved. The crack type output by the model guides subsequent quantification strategies, forming a "classification-measurement" pipeline. If the model determines a surface crack, subsequent measurements only require measuring its length and width; if it determines an internal defect, parameters such as its depth and area must be calculated. This close connection between these steps makes the entire inspection process more efficient and organized.

[0063] Furthermore, the crack detection results are quantitatively analyzed based on the output, and warning information is issued according to a preset threshold, including: measurement of crack length and width and calculation of curvature, calculation of internal defect area of ​​the crack and analysis of thermal diffusivity.

[0064] Surface crack quantification primarily involves length and width measurement and curvature calculation. Based on sub-pixel edge coordinates from the visual image, the actual length and width are calculated using pixel density (0.3mm / pixel), with an error of ≤0.1mm + 5% L. Sub-pixel edge coordinates can more precisely determine the crack boundary, and pixel density can be used to convert pixel coordinates into actual length and width, resulting in more accurate measurement results. Furthermore, controlled error ranges ensure the reliability of measurement results, providing more accurate data for subsequent evaluation and processing.

[0065] Curvature calculation involves cubic spline interpolation of the crack edge points. The curvature radius is calculated for every 10mm segment. A curvature greater than 100mm-1 is considered a sharp crack (high risk). Curvature reflects the degree of curvature of the crack. Sharp cracks typically indicate faster crack expansion and greater damage to the building structure. By calculating curvature, the risk level of the crack can be assessed, providing a basis for timely action.

[0066] Internal defect assessment mainly includes the calculation of defect area and analysis of thermal diffusivity. Ultrasonic TFM imaging results are used to calculate the defect area (number of pixels × 0.5mm 2 / pixel), area>500mm 2 It is determined to be a serious defect. Ultrasonic TFM imaging can provide detailed images of internal defects of buildings. By counting the number of pixels of defects in the image and combining the actual area corresponding to the pixels (0.5mm 2 / pixel), the area of ​​the defect can be accurately calculated. Judging the severity of the defect based on the area size helps to detect potential safety hazards in a timely manner.

[0067] Combined with the infrared thermal diffusivity α, if α is less than the average value of the normal area, it indicates that the defect may be accompanied by moisture infiltration (further radar testing is required). Thermal diffusivity reflects the thermal conductivity of the material. When the thermal diffusivity of the defect area is significantly lower than that of the normal area, it indicates that the thermal conductivity of the area has changed, possibly due to moisture infiltration. Moisture infiltration can accelerate the expansion of cracks and damage the building structure, so further radar testing is necessary to determine the extent and scope of moisture infiltration.

[0068] The warning logic uses a multi-level threshold setting, which is divided into level 1 warning and level 2 warning. Level 1 warning (yellow): surface crack width > 0.2mm or internal defect depth > 30mm; Level 2 warning: surface crack width > 0.3mm and length > 1m, or internal defect area > 1000mm 2 The multi-level threshold setting can provide different levels of early warning according to the severity of the cracks, allowing users to understand the safety status of the building structure in a timely manner and take corresponding measures.

[0069] Warning information includes: detection time, location coordinates (X, Y, Z), defect parameters, and historical comparison data (such as the crack width growth rate in the previous detection in the area). This information helps users fully understand the crack situation, including the crack's occurrence time, location, size, and development trend. Historical comparison data can reflect crack changes, such as the crack width growth rate. A rapid growth rate indicates a dangerous crack development trend and requires timely treatment.

[0070] The warning threshold combines building regulations with historical data to make the warning more scientific and reasonable. Building regulations have clear provisions for parameters such as crack width and depth. Historical data can reflect the development trend of cracks. By combining this information, a more reasonable warning threshold can be set. If a high-risk defect is detected (such as a level 2 warning), the system automatically triggers the robotic arm to return to the area and conduct a re-inspection with higher precision (the visual shooting distance is shortened to 0.5m, and the ultrasonic probe frequency is increased to 10MHz), forming a closed-loop detection. This re-inspection mechanism can ensure the accuracy of the test results, avoid missed detections and misjudgments, and provide more reliable protection for the safety of building structures.

[0071] A method for using a crack detection device for construction engineering includes the following steps: a first step of obtaining multimodal nondestructive testing information of a building structure;

[0072] The second step is to preprocess the multimodal nondestructive testing information obtained by the data acquisition module to obtain a standardized multimodal testing data set;

[0073] Step 3: Based on the input cross-modal feature fusion tensor, cracks are identified and classified based on the improved deep learning model, and the crack detection results are output;

[0074] Step 4: Based on the output crack detection results, perform quantitative analysis on the crack detection results and issue early warning information based on the preset threshold.

[0075] The above are only embodiments of the present invention, and the circuits, electronic components and modules involved are all prior art, which can be fully implemented by those skilled in the art. It is needless to say that the content protected by this application does not involve improvements to software and methods. Common knowledge such as the specific structures and characteristics known in the scheme are not described in detail here. Ordinary technicians in the field are aware of all common technical knowledge in the technical field of the invention before the application date or priority date, can obtain all prior art in the field, and have the ability to apply conventional experimental means before that date. Ordinary technicians in the field can improve and implement this scheme in combination with their own abilities under the inspiration given by this application. Some typical known structures or known methods should not become obstacles for ordinary technicians in the field to implement this application. It should be pointed out that for those skilled in the art, without departing from the structure of the present invention, several variations and improvements can be made, which should also be regarded as the scope of protection of the present invention. These will not affect the effect of the implementation of the present invention and the practicality of the patent.

Claims

1. A crack detection device for construction engineering, characterized by: include, A data acquisition module configured to acquire multimodal nondestructive testing information of a building structure; a data preprocessing module configured to preprocess the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set; a data processing module configured to match and fuse different modal data in the standardized multimodal detection dataset in time and space, and to associate and integrate feature information of different modal data in the standardized multimodal detection dataset to obtain a cross-modal feature fusion tensor; The crack identification and classification module is configured to identify and classify cracks based on the input cross-modal feature fusion tensor based on the improved deep learning model and output the crack detection results; The intelligent early warning module is configured to perform quantitative analysis on the crack detection results according to the output crack detection results, and issue early warning information according to a preset threshold value.

2. A crack detection device for construction engineering according to claim 1, characterized in that: The data acquisition module includes a visual image acquisition module for acquiring image information of the building structure surface at the same moment, an ultrasonic signal acquisition module for acquiring internal structure information of the building structure at the same moment, an infrared data acquisition module for acquiring temperature information of the building structure surface at the same moment, and a synchronous triggering module. The multimodal nondestructive testing information includes visual images, ultrasonic signals, and temperature information of the building structure surface.

3. A crack detection device for construction engineering and a method of use according to claim 2, characterized in that: The synchronous trigger module includes a hardware timer, the visual image acquisition module includes an industrial camera, the ultrasonic signal acquisition module includes a phased array ultrasonic probe, and the infrared data acquisition module includes an infrared thermal imager. The hardware timer is electrically connected to the industrial camera, the phased array ultrasonic probe, and the infrared thermal imager.

4. A crack detection device for construction engineering according to claim 2, characterized in that: The preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes: Visual image preprocessing first uses bilateral filtering to denoise the image, then performs morphological opening operation to remove tiny noise points, then uses Frangi vascular filtering algorithm to enhance crack features, and finally uses adaptive threshold segmentation to extract suspected crack area ROI.

5. A crack detection device for construction engineering according to claim 4, characterized in that: The preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes: Ultrasonic signal processing includes three steps: time domain filtering, phase correction, and feature parameter extraction. Specifically, time domain filtering is to perform preliminary purification of the ultrasonic signal, and a 50-5000Hz bandpass filter is applied to remove environmental noise; phase correction is to eliminate the phase offset caused by the coupling difference between the probe and the surface through Hilbert transform de-envelope; feature parameter extraction is to extract the arrival time t of the defect echo from the ultrasonic signal. p , peak intensity A p , rise time t r， Construct ultrasound feature vector [t p ,A p ,t r ].

6. A crack detection device for construction engineering according to claim 5, characterized in that: The preprocessing of the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set includes: Infrared data processing includes two steps: temperature normalization and Gaussian pyramid downsampling. Specifically, the collected temperature value is converted into the difference relative to the ambient temperature (ΔT = TT env ), improve the visual recognition of temperature distribution through pseudo-color mapping; use Gaussian blur or median filtering to smooth noise, segment abnormal hot areas according to temperature thresholds, and locate the detection area in combination with the building structure model.

7. A crack detection device for construction engineering according to claim 6, characterized in that: The data processing module includes a spatiotemporal registration module configured to obtain the external parameters of the visual camera, including the rotation matrix R and the translation vector T, and convert the pixel coordinates in the visual image into points in the camera coordinate system; obtain the position information of the ultrasound probe, and calculate the position information of the ultrasound probe by the formula P world =R·P camera +T=R′·P ultrasonic +T′ unifies the visual and ultrasound data into the world coordinate system, where P camera is the camera coordinate system point converted from visual pixel coordinates, P ultrasonic is the ultrasound scanning coordinate.

8. A crack detection device for construction engineering according to claim 7, characterized in that: The data processing module includes a feature extraction and association module, which is configured to extract edge, texture, and color features from the visual image; extract the arrival time, peak intensity, and rise time features of the defect echo from the ultrasonic signal; extract the temperature difference and thermal diffusivity features from the infrared data; and for each visual pixel point (x, y) belonging to the crack ROI, associate it with the ultrasonic feature (t p ,A p ) and infrared features (ΔT, thermal diffusivity α) to construct a three-dimensional feature vector [g(x,y),u(x,y),i(x,y)], and obtain a cross-modal feature fusion tensor, where g(x,y) is the visual feature, u(x,y) is the ultrasonic feature, and i(x,y) is the infrared feature. In order to highlight the key modal features, the attention mechanism is used to assign feature weights: w=σ(W g g+W u u+W i i), where σ is the Sigmoid function, W g 、W u 、W i are the weight matrices of visual, ultrasonic and infrared features respectively.

9. A crack detection device for construction engineering according to claim 8, characterized in that: The crack recognition and classification module is configured to identify and classify cracks based on the input cross-modal feature fusion tensor and an improved deep learning model, and output crack detection results. This includes encoding the fused feature vector [g,u,i] into a pseudo-color image (R channel = visual probability, G channel = ultrasonic echo intensity, B channel = infrared ΔT) as the input of the YOLOv8 model. At the same time, an auxiliary input branch is added to directly input the ultrasonic C-scan grayscale image and infrared thermal image to form a multi-branch network structure. A cross-modal feature interaction module is introduced into the neck network (PAFPN): Among them F vis is the visual feature, F ult For ultrasonic features, the defect correlation in visual features is enhanced by channel-level attention, and the YOLOv8 model outputs crack type, confidence, and geometric parameters for each detection box.

10. A method for using a crack detection device for construction engineering, characterized in that: The following steps are included: The first step is to obtain multimodal nondestructive testing information of the building structure; Step 2; Preprocessing the multimodal nondestructive testing information acquired by the data acquisition module to obtain a standardized multimodal testing data set; Step 3; According to the input cross-modal feature fusion tensor, cracks are identified and classified based on the improved deep learning model, and crack detection results are output; Step 4: Based on the output crack detection results, perform quantitative analysis on the crack detection results and issue early warning information based on the preset threshold.

Images