Bridge steel structure welding quality detection system and method

Abstract

The invention belongs to the technical field of bridge steel structures, and discloses a bridge steel structure welding quality detection system and method, and the system comprises a data collection module which collects related data of welding; the route planning module is used for planning a cruise route of the unmanned aerial vehicle; the inspection module performs cruise detection on the welding quality of the steel structure; the image preprocessing module is used for preprocessing the image; the feature extraction module is used for carrying out feature extraction on the preprocessed image; the abnormity recognition module recognizes that welding abnormity exists; the infrared data processing module is used for identifying the welding problem of the steel structure; and the data fusion module is used for combining the welding abnormal condition recognized by the abnormal recognition module with the welding quality result detected by the infrared data processing module, and recognizing the welding quality problem of the steel structure. According to the system, automatic inspection can be carried out, the detection speed is increased, visual detection and infrared detection results are combined through the data fusion module, and the welding quality problem recognition accuracy is improved.

Description

Technical Field

[0001] This application relates to the technical field of bridge steel structures, and more specifically, to a welding quality detection system and method for bridge steel structures. Background Art

[0002] In the complex process of bridge construction, steel structure welding is a crucial link, which directly determines the overall strength, stability and durability of the bridge. A steel structure, as a structural system constructed from steel materials, mainly consists of profiled steel (such as I-beams, angle steels, channel steels, etc.), steel plates and other steel components (such as steel beams, steel columns, steel trusses, etc.). These components are finally assembled into the main skeleton of the bridge through precise cutting, shaping and assembly. Welding, as the main means of connecting steel structures, its quality is directly related to the safety and service life of the bridge. Therefore, during and after the welding process, strict inspection and control of the welding quality must be carried out.

[0003] The prior art document with the publication number CN20251823U provides a non-destructive testing device for the welding quality of bridge steel structures, including a workbench, a support frame fixedly connected to the top of the workbench, a conveyor fixedly installed on the top of the workbench, and limit mechanisms arranged on both the front and rear sides of the conveyor. Through the cooperation of a motor, a threaded rod, a threaded plate, a first electric push rod, an ultrasonic detector, an ultrasonic detection probe and an X-ray detector, this utility model can effectively perform non-destructive testing on the steel structure, so that during the testing, various methods can be used to detect the damage condition of the steel structure, thereby increasing the practicability of the device. Through the cooperation of a fixed cover, a movable slide rod, an installation slider, a second electric push rod, an L-shaped plate, a spring and a limit push plate, this utility model can limit the steel structure, so that when the steel structure is conveyed, the position deviation of the steel structure can be prevented, enabling the device to effectively perform non-destructive detection on the steel structure.

[0004] Although the above prior art solutions can achieve relevant beneficial effects through the structures of the prior art, there are still the following defects: 1. Sometimes the height of the bridge steel structure is very high, and quality inspectors cannot quickly and easily perform welding quality inspection. Performing welding quality inspection manually has low work efficiency and is time-consuming and laborious. 2. The accuracy of welding quality inspection by manual is not high, and welding problems such as false welding and cavities cannot be accurately detected.

[0005] In view of this, we propose a welding quality detection system and method for bridge steel structures. Summary of the Invention

[0006] 1. Technical Problems to be Solved

[0007] The purpose of this application is to provide a bridge steel structure welding quality detection system and method, which solves the technical problems raised in the above-mentioned background technology, realizes automatic inspection using drones, reduces the need for manual inspection, improves the inspection speed, and improves work efficiency; through the data fusion module, the visual inspection and infrared inspection results are combined, and the technical effect of improving the accuracy of identifying welding quality problems is achieved.

[0008] 2. Technical solution

[0009] The technical solution of this application provides a bridge steel structure welding quality detection system, including:

[0010] Data acquisition module: Collect a large amount of relevant data (including drawings and welding requirements) of bridge steel structure welding, including information such as welding positions, welding methods, and weld bead thicknesses; include images of normal welding and welding with abnormal conditions, and label the images as reference samples; the images of welding abnormal conditions include welding problems such as cracks, welding deformation, lack of fusion, porosity, arc pits, and incomplete fusion; establish a reference sample database;

[0011] Route planning module: Based on the bridge structure and detection requirements, use algorithms to optimize the drone cruise path and plan the drone's cruise route; according to the actual detection situation, adjust the drone cruise route in real time.

[0012] Inspection module: Includes drones, on which a high-definition camera, GPS locator, and infrared welding light transmittance detector are set; conducts cruise detection on the welding quality of the bridge's steel structure;

[0013] Image preprocessing module: Preprocess the collected images, including filtering and denoising, grayscale conversion, image segmentation, and image enhancement. After obtaining the solder joint images processed by image enhancement and filtering, use computer image processing technology to segment the solder joint area and the background area, search and locate each solder joint one by one, and obtain information about the boundaries, center positions, and other shape information of the solder joint area, so as to extract certain shape and position characteristics of this area, thereby realizing the recognition of specific targets.

[0014] Feature extraction module: Extract features from the preprocessed images, extract features related to welding abnormal conditions, including welding problems such as cracks, welding deformation, lack of fusion, porosity, arc pits, and incomplete fusion; based on the labeled images, construct a feature library of welding defects. Automatically extract the features of welding quality problems such as cracks, deformation, lack of fusion, porosity, arc pits, and incomplete fusion in the images.

[0015] Abnormality recognition module: Compare and analyze the image after feature extraction with the reference sample to identify abnormal welding conditions, including welding problems such as cracks, welding deformation, lack of fusion, porosity, craters, and incomplete fusion; identify the size and location of the welding abnormalities.

[0016] Infrared data processing module: Process and analyze the data detected by the infrared welding light transmittance detector to identify welding problems in the steel structure.

[0017] Data fusion module: Combine the welding abnormalities identified by the abnormality recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure.

[0018] Alarm module: Includes an alarm. When welding quality problems are detected, an alarm is issued and the location information of the welding quality problems is recorded.

[0019] PLC control module, which is network-connected to the route planning module, the inspection module, the data fusion module, and the alarm module, and is network-connected to the remote control center through a wireless signal.

[0020] As an alternative solution of the present invention, the feature extraction module uses NSST to decompose the welding defect image, and uses PCNN to extract the approximate area of the defect for the low-frequency component that reflects the rough approximation of the defect; then, perform inverse NSST on the low-frequency component and the high-frequency component after background suppression to obtain the high-frequency feature image. After rough segmentation of it, use the improved CV model to optimize the contour of the defect to obtain the fine edge of the defect; finally, fuse the extraction results to obtain the finally extracted defect. The specific steps are as follows:

[0021] Step1: Non-subsampled Shearlet transform, perform single-layer non-subsampled Shearlet decomposition on the welding image I to obtain the low-frequency component L and six high-frequency components H i , perform non-subsampled Laplacian pyramid decomposition using the "Maxflat" filter, and perform direction analysis using the "Meyer" window function.

[0022] Step2: Use the low-frequency image for rough segmentation of the defect. After NSST decomposition, the low-frequency component has less noise content, but its defect contour is relatively blurred. Use it to extract the main area information of the defect to reduce noise interference.

[0023] Take the low-frequency component L as the input of the pulse coupled neural network (PCNN), and use Shannon entropy to determine the optimal number of iterations T opt . After PCNN processing, obtain the roughly segmented welding defect image D coarse .

[0024] Step3: Low-frequency image background suppression. Determine the threshold T according to the valley inflection point of the image histogram bg , and suppress the background part. Use a spline curve to fit the histogram to reduce the influence of local fluctuations on threshold selection. Weld defect images usually have low contrast and large background fluctuations, while the gray values of the weld bead part are generally large. The corresponding threshold can be found according to a valley inflection point of the image histogram curve, and the fluctuating background part is directly set to zero to reduce the interference of the background on defect extraction. To prevent misselection of the threshold caused by local fluctuations of the histogram curve, a spline curve is selected to fit the histogram curve.

[0025] Step4: High-frequency feature image construction. Use the NSST inverse transform to reconstruct the low-frequency image L bg-inh after background suppression and the high-frequency image H i to obtain the high-frequency feature image H feature .

[0026] Step5: Use the high-frequency feature image for defect fine segmentation. Perform threshold segmentation on H feature to obtain the initial segmentation result. Use the obtained image as the initial condition of the improved CV model to continuously approximate the optimal edge of the defect. The evolution equation of the level set function φ is as follows

[0027]

[0028] ω(x) = {ω1, φ(x) > 0; ω2, φ(x) < 0};

[0029] In the formula, z represents the image gray value; c1 and c2 are the average gray values of the image in the internal and external regions of the curve C; μ, λ1, λ2 are weight parameters, δ is the Dirac function. φ is the level set function. ω1 increases the height of the 0-level set plane, making the boundary move in the direction of larger gray values, while ω2 is used to reduce the height of the 0-level set plane, making the boundary move in the direction of smaller gray values. t is the time variable; represents the change rate of the level set function φ with respect to time t, that is, it describes how the level set evolves over time.

[0030] Step6: Fuse the results obtained by high-frequency component segmentation. Perform logical AND or OR operations on the defect image D high extracted from the high-frequency feature image and the defect image D low extracted from the low-frequency image to obtain the final defect image D final . Use the Sobel operator to extract the fine edges of the defects in D final . On the one hand, it can extract the fine edges of the defects, and on the other hand, it can effectively remove background noise.

[0031] In this technical solution, the feature extraction module can effectively extract detailed information of welding defects from welding images, including the approximate area, fine edges, and contours of the defects. This method combines the multi-scale and multi-directional decomposition capabilities of NSST and the edge optimization ability of the improved CV model, improving the accuracy and reliability of welding defect detection.

[0032] As an alternative solution of the present invention, the infrared data processing module processes and analyzes the detected data to identify welding problems of steel structures.

[0033] As an alternative solution of the present invention, the route planning module optimizes the UAV cruise path based on the bridge structure and detection requirements using the A* algorithm, including the following steps:

[0034] 1. Recording of bridge numbers and geographical location information: Different bridges are assigned unique numbers for subsequent management and identification. Record geographical location information such as the longitude, latitude, and altitude of each bridge, which helps the UAV for positioning and navigation.

[0035] 2. Bridge structure analysis: Collect structural data such as CAD drawings and 3D models of the bridge to understand the overall layout of the bridge, the positions and dimensions of key components. Define the purpose, scope, and accuracy requirements of the detection, and determine the key areas and details of the bridge to be detected.

[0036] 3. UAV cruise path planning:

[0037] Environmental modeling: Based on the geographical location information and structural data of the bridge, establish an environmental model, including the bridge structure, surrounding environment (such as buildings, trees, power lines, etc.), and flight restricted areas (such as no-fly zones, restricted heights, etc.).

[0038] Path search: Use the A algorithm to search for paths in the environmental model. The A algorithm selects the next moving direction by comprehensively considering the actual cost and heuristic estimated cost of the path to achieve the goal of finding the optimal path. Set the starting point (UAV takeoff point), ending point (bridge detection point), and obstacles (bridge structure, obstacles in the surrounding environment). Design a cost function according to the detection requirements and environmental factors, such as flight time, flight distance, obstacle avoidance cost, etc. The A* algorithm will continuously search and update the path from the starting point to the ending point until the optimal path is found or it is determined that there is no feasible path.

[0039] Path optimization: Smooth the optimal path found by the A* algorithm to reduce unnecessary turns and undulations, improving the flight stability and safety.

[0040] 4. Cruise route verification and adjustment:

[0041] Simulation verification: Use simulation software to simulate the UAV cruising along the planned path to verify the feasibility and effectiveness of the path. According to the results of the simulation verification, make necessary adjustments to the path to adapt to different bridge structures and inspection requirements.

[0042] As an alternative solution of the present invention, environmental modeling uses remote sensing technologies such as lidar (LiDAR) and UAV aerial photography to obtain three-dimensional data of the bridge and its surrounding environment. Use point cloud processing software (such as PCL, CloudCompare, etc.) to process the obtained three-dimensional data, and extract the structural information of the bridge and the obstacle information of the surrounding environment. Import the processed data into Blender modeling software to automatically generate an environmental model. Specifically, it includes the following steps:

[0043] 1. Data acquisition: Use LiDAR equipment to scan the bridge and its surrounding environment from different angles to obtain high-precision three-dimensional point cloud data. Deploy a UAV equipped with a high-definition camera and / or LiDAR sensor to conduct aerial photography of the bridge and collect images and three-dimensional data.

[0044] 2. Point cloud preprocessing: Synchronize the LiDAR point cloud data and the UAV aerial photography data in time and register them spatially. Use point cloud processing software (such as PCL, CloudCompare) to remove noise and outliers in the point cloud.

[0045] 3. Structural information extraction: Separate the bridge structure points from the background points according to the attributes of the point cloud (such as reflection intensity, height, density). Identify the key structural features of the bridge through feature extraction, such as beams, columns, nodes, etc.

[0046] 4. Obstacle detection: Analyze obstacles in the surrounding environment, such as buildings, trees, terrain, etc. Mark the obstacles that may affect bridge inspection and maintenance in the point cloud data.

[0047] 5. Data processing and modeling: Optimize the extracted structural information and obstacle information to remove redundant data. Import the processed data into modeling software such as Blender to automatically generate a three-dimensional model of the bridge and its surrounding environment.

[0048] 6. Model refinement: Add details of the bridge structure, such as textures, materials, etc. in the three-dimensional model. Simulate the influence of environmental factors on the bridge, such as lighting, shadows, etc.

[0049] 7. Model verification and application: Verify the accuracy and reliability of the model by comparing it with the actual bridge. Use the generated environmental model for bridge structural analysis, health monitoring, maintenance planning, etc.

[0050] As an alternative solution of the present invention, the anomaly recognition module uses Euclidean distance and cosine similarity to measure the difference between two vectors in feature matching. Combine the formula with welding problems such as cracks, welding deformation, lack of fusion, porosity, crater, and incomplete fusion to identify welding quality problems, according to the following steps:

[0051] Feature vectorization: First, convert the welding defects (such as cracks, deformation, porosity, crater, incomplete fusion, etc.) identified in the image into numerical feature vectors. These features may include the size, shape, texture, location, etc. of the defects.

[0052] Size feature: Measure the size features such as the width and length of defects such as cracks and porosity.

[0053] Shape feature: Extract the shape descriptors of the defects, such as roundness, rectangularity, etc.

[0054] Texture feature: Analyze the texture features of the defect area, such as roughness, uniformity, etc.

[0055] Location feature: Record the coordinate position of the defect in the welding area.

[0056] Establish a reference sample library: Create an image database containing normal welds and various known welding defects, and extract feature vectors for these images as reference samples. Collect images of normal welds and their feature vectors. Collect defect samples, including welding images with defects such as cracks, welding deformation, lack of fusion, porosity, crater, and incomplete fusion and their feature vectors.

[0057] Feature matching process: For a new welding image, extract similar feature vectors, and then use the following formula to calculate the difference between the feature vectors of the new image and the reference samples:

[0058]

[0059] Cosine Similarity=(A·B) / (||A||||B||);

[0060] In the formula, d(A,B) is the Euclidean distance between vectors A and B. A and B are two feature vectors representing the features in the welding image. Ai and Bi are the values of the i-th feature in vectors A and B respectively. n is the total number of features in feature vectors A and B. Cosine Similarity is the cosine similarity between vectors A and B, representing the cosine value of the angle between them. A·B is the dot product of vectors A and B. ||A|| and ||B|| are the norms of vectors A and B (usually the Euclidean norm), that is, the lengths of the vectors.

[0061] The Euclidean distance is used to measure the straight-line distance between feature vectors. The smaller the distance, the more similar it indicates. The cosine similarity measures the angle between feature vectors. The closer the cosine value is to 1, the more similar it indicates.

[0062] Calculating through formulas helps us quantify the differences between feature vectors, thereby identifying welding defects. Feature vectors A and B may contain various features extracted from welding images, such as the width and length of cracks, the angle of welding deformation, the texture of the non-welded area, etc. By comparing these feature vectors with the feature vectors of reference samples (welding images of normal welding or known defects), we can evaluate the welding quality and identify potential welding problems.

[0063] Threshold setting: Set a threshold to distinguish whether the welding quality is qualified. If the calculated distance or similarity exceeds this threshold, it is considered that there is a problem with the welding.

[0064] Set a distance threshold Td. If it exceeds this threshold, it is considered that there is a welding defect.

[0065] Set a similarity threshold Tc. If it is lower than this threshold, the welding quality is considered unqualified.

[0066] Defect identification: Based on the results of feature matching, identify the types of defects in the welding image. For example, if the similarity between a certain feature vector and the feature vector of cracks in the database is very low, it is considered that there are no cracks in this welding.

[0067] Size and position location: For the identified welding defects, use image processing techniques to determine their specific positions and sizes in the image, providing information for subsequent repairs or further analysis.

[0068] Boundary detection: Use image processing techniques and Canny, Sobel or other edge detection algorithms to identify the boundaries of defects in the image. Convert the image into a binary image to facilitate clear identification of the boundaries. Implement through an automatic boundary tracking algorithm to determine the defect contour.

[0069] Size measurement: Calculate the size of the defect, such as length, width, etc.

[0070] Defect area calculation: Calculate the number of pixels in the binarized defect area and convert it into the actual area. Adefect = Pixel Count × Pixel Area where Adefect is the defect area, Pixel Count is the number of pixels in the defect area, and Pixel Area is the actual area corresponding to each pixel.

[0071] Length and width measurement: For regularly shaped defects, directly measure their length and width; for irregularly shaped ones, use methods such as the minimum enclosing rectangle or fitting ellipse.

[0072] L = max(x) - min(x);

[0073] W = max(y) - min(y);

[0074] L and W are the length and width of the defect.

[0075] Position determination: Record the specific coordinate position of the defect in the welding image.

[0076] As an alternative embodiment of the present invention, the data fusion module combines the welding anomalies identified by the anomaly recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure, including the following steps;

[0077] Data preprocessing: Clean the data to remove invalid or incorrect data points. Convert data in different formats into a unified format for easy processing. Normalize the data to eliminate the influence of different dimensions.

[0078] Data fusion processing:

[0079] Feature-level fusion: F combined = w1F visual + w2F IR ; where F combined is the fused feature vector, F visual and F IR are the feature vectors from the anomaly recognition module and the infrared data processing module respectively, and w1 and w2 are the corresponding weights.

[0080] Welding quality problem identification: Quality Issue = {[Yes, if

[0081] Fcombined < T or D = Reject]; [No, otherwise]}, where Quality Issue is a boolean variable indicating whether there is a welding quality problem, T is a preset threshold, and Reject is the result of decision-level fusion.

[0082] Classification and quantification: C = h(Q); where C is the classification result of the welding quality problem, h is the classification function, and the problem is classified into different categories and severities according to the value of Q.

[0083] Result output: Output the evaluation results of the data fusion module, including the overall evaluation of the welding quality, the detailed information of the problem area, and the recommended repair measures.

[0084] In this technical solution, the data fusion module can integrate the results of two independent modules to provide a more comprehensive and accurate assessment of welding quality. This method can improve the reliability and efficiency of the welding quality monitoring system, helping operators quickly identify and handle potential welding problems.

[0085] As an alternative solution of the present invention, an angle position adjustment mechanism is provided on the unmanned aerial vehicle;

[0086] The angle position adjustment mechanism includes an electric telescopic rod, an electric turntable and a motor;

[0087] An electric telescopic rod is fixedly arranged on the unmanned aerial vehicle; an electric turntable is fixedly arranged on the movable rod of the electric telescopic rod;

[0088] The output end of the electric turntable is fixedly provided with a motor, the output end of the motor is coaxially fixedly provided with a shaft rod, and a high-definition camera is fixedly arranged on the shaft rod;

[0089] Through the above technical solution, starting the electric telescopic rod drives the electric turntable, the motor and the high-definition camera to move, adjusting the position where the high-definition camera extends out; driving the motor and the high-definition camera to rotate through the electric turntable, and the motor drives the high-definition camera to rotate in another direction, realizing the adjustment of the angle of the high-definition camera, so that the high-definition camera is aligned with the welding position, facilitating the acquisition of high-definition images. Especially when detecting the welding at the corners or the upper end of the top, adjusting the angle of the high-definition camera can acquire high-definition images.

[0090] As an alternative solution of the present invention, a welding defect marking mechanism is provided on the unmanned aerial vehicle. The welding defect marking mechanism includes a welding defect marking liquid storage tank, a suction pump and a spray pipe;

[0091] A welding defect marking liquid storage tank is fixedly arranged on the unmanned aerial vehicle, and a suction pump is fixedly arranged on the welding defect marking liquid storage tank; the input end of the suction pump is communicated with the inside of the welding defect marking liquid storage tank through a pipeline; the welding defect marking liquid storage tank is filled with red welding defect marking liquid, and the welding defect marking liquid can be selected as red or yellow according to needs. The welding defect marking liquid can also be red or yellow spray paint.

[0092] A spray pipe is fixedly arranged on the shaft rod; the spray pipe is communicated with the output end of the suction pump through a hose;

[0093] In this technical solution, when a welding quality defect is detected on the bridge steel structure, the welding defect marking liquid in the welding defect marking liquid storage tank is sucked into the spray pipe through the suction pump and then sprayed near the welding defect, facilitating quickly finding the position of the welding defect during later maintenance.

[0094] The present invention provides a method for detecting the welding quality of bridge steel structures, including the following steps:

[0095] S1. The data acquisition module collects a large amount of data related to the welding of bridge steel structures (including drawings and welding requirements), including information such as the welding position, welding method, and weld bead thickness; includes images of normal welding and abnormal welding conditions, and annotates the images as reference samples;

[0096] S2. The route planning module plans the cruise route of the drone based on the bridge structure and inspection requirements, and optimizes the drone's cruise path using an algorithm;

[0097] S3. The inspection module conducts cruise inspections on the welding quality of the bridge steel structure through the drone;

[0098] S4. The image preprocessing module preprocesses the collected images, including filtering and denoising, grayscale conversion, image segmentation, and image enhancement. The feature extraction module extracts features from the preprocessed images, and extracts features related to abnormal welding conditions, including welding problems such as cracks, welding deformation, lack of fusion, porosity, craters, and lack of penetration;

[0099] S5. The abnormal recognition module compares and analyzes the images after feature extraction with the reference samples, and identifies abnormal welding conditions, including welding problems such as cracks, welding deformation, lack of fusion, porosity, craters, and lack of penetration; identifies the size and position of the abnormal welding conditions;

[0100] S6. The infrared data processing module processes and analyzes the data detected by the infrared welding light transmittance detector to identify welding problems in the steel structure;

[0101] S7. The data fusion module combines the welding abnormal conditions identified by the abnormal recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify welding quality problems in the steel structure;

[0102] S8. When welding quality problems are detected, the alarm module issues an alarm and records the location information of the welding quality problems. By starting the suction pump, the welding defect marking liquid in the welding defect marking liquid storage tank is sucked into the nozzle and then sprayed near the welding defect, so that the welding defect location can be quickly found during later maintenance.

[0103] 3. Beneficial effects

[0104] One or more technical solutions provided in the technical solution of the present application have at least the following technical effects or advantages:

[0105] 1. The present invention uses a drone for automated inspection, reducing the need for manual inspection, improving the inspection speed, and improving work efficiency.

[0106] 2. The visual detection and infrared detection results are combined through the data fusion module, improving the accuracy of identifying welding quality problems.

[0107] 3. By using the welding defect marking liquid, this method can visually mark the defect positions on the physical structure, facilitating subsequent repairs. The precise positioning and marking of defects reduce the time for maintenance personnel to search for and identify problems, improving the maintenance efficiency. The welding quality detection method for bridge steel structures not only improves the detection accuracy and efficiency but also provides strong technical support for the maintenance and management of bridges.

[0108] 4. The extension length and angle of the high-definition camera are adjusted through the angle position adjustment mechanism, aligning the high-definition camera with the welding position to facilitate the acquisition of high-definition images. Especially when detecting the welding at the corners or the upper end of the top, adjusting the angle of the high-definition camera can acquire high-definition images, expanding the application range. BRIEF DESCRIPTION OF THE DRAWINGS

[0109] Figure 1 It is the overall schematic diagram of the bridge steel structure welding quality detection system disclosed in a preferred embodiment of the present application;

[0110] Figure 2 It is the schematic diagram of the angle position adjustment mechanism and the welding defect marking mechanism of the bridge steel structure welding quality detection system disclosed in a preferred embodiment of the present application;

[0111] Figure 3 It is the schematic flow diagram of the feature extraction module using the improved CV model to extract defects.

[0112] REFERENCE SIGNS:

[0113] 1. Drone; 2. Electric telescopic rod; 3. Electric turntable; 4. Motor; 5. High-definition camera; 6. Positioning ring; 7. Welding defect marking liquid storage tank; 8. Suction pump; 9. Shaft rod; 10. Spray pipe. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0114] The following further describes the present application in detail with reference to the accompanying drawings of the specification.

[0115] Referring to Figure 1 , the embodiment of the present application provides a bridge steel structure welding quality detection system, including:

[0116] Data acquisition module: A large amount of data related to the welding of bridge steel structures (including drawings and welding requirements) is collected, including information such as the welding position, welding method, and weld bead thickness; images of normal welding and welding with abnormal conditions are included and annotated as reference samples; the images of welding with abnormal conditions include welding problems such as cracks, welding deformation, lack of fusion, porosity, crater, and incomplete fusion; a reference sample database is established;

[0117] Route planning module: Based on the bridge structure and inspection requirements, use algorithms to optimize the UAV cruise path and plan the UAV cruise route; adjust the UAV cruise route in real time according to the actual inspection situation.

[0118] Inspection module: Includes a UAV equipped with a high-definition camera, a GPS locator, and an infrared welding light transmittance detector; conducts cruise inspections on the welding quality of the steel structure of the bridge; the infrared welding light transmittance detector is an instrument specifically used to detect welding quality. Using infrared spectroscopy technology, it determines whether the welding quality is qualified by measuring the transmittance of the material in the infrared spectral region.

[0119] Image preprocessing module: Preprocesses the collected images, including filtering and denoising, grayscale conversion, image segmentation, and image enhancement. Applies Gaussian filtering, median filtering, or other denoising algorithms to reduce random noise in the images. After obtaining the solder joint images processed by image enhancement and filtering, uses computer image processing technology to segment the solder joint area and the background area, searches and locates each solder joint one by one, obtains information about the boundaries, center positions, and other shape-related information of the solder joint area, and extracts certain shape and position characteristics of this area to achieve the recognition of specific targets.

[0120] Feature extraction module: Extracts features related to welding abnormalities from the preprocessed images, including welding problems such as cracks, welding deformation, false soldering, pores, craters, and lack of fusion; constructs a feature library of welding defects based on the annotated images. Automatically extracts features of welding quality problems such as cracks, deformation, false soldering, pores, craters, and lack of fusion in the images. Uses the Sobel operator for edge detection; calculates features of the gray-level co-occurrence matrix (GLCM), such as energy, contrast, uniformity, and directionality, etc.; calculates shape descriptors, such as circularity, rectangularity, etc., to describe the shape characteristics of the welding area.

[0121] Abnormality recognition module: Compares and analyzes the images after feature extraction with reference samples to identify welding abnormalities, including welding problems such as cracks, welding deformation, false soldering, pores, craters, and lack of fusion; identifies the size and location of the welding abnormalities.

[0122] Infrared data processing module: Processes and analyzes the data detected by the infrared welding light transmittance detector to identify welding problems in the steel structure.

[0123] Data fusion module: Combines the welding abnormalities identified by the abnormality recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure.

[0124] Alarm module: includes an alarm. When welding quality problems are detected, it issues an alarm and records the location information of the welding quality problems;

[0125] PLC control module, which is network-connected to the route planning module, the inspection module, the data fusion module and the alarm module, and is network-connected to the remote control center through wireless signals.

[0126] Refer to Figure 3 , Feature extraction module. It uses NSST to decompose the welding defect image, and uses PCNN to extract the approximate area of the defect from the low-frequency component that reflects the rough approximation of the defect; Then, perform inverse NSST on the background-suppressed low-frequency component and high-frequency components to obtain the high-frequency feature image. After rough segmentation of it, use the improved CV model to optimize the contour of the defect to obtain the fine edge of the defect; Finally, fuse the extraction results to obtain the finally extracted defect. The specific steps are as follows:

[0127] Step1: Non-subsampled Shearlet transform. Perform single-layer non-subsampled Shearlet decomposition on the welding image I to obtain the low-frequency component L and six high-frequency components H i , Use the "Maxflat" filter for non-subsampled Laplacian pyramid decomposition, and use the "Meyer" window function for direction analysis.

[0128] Step2: Coarse segmentation of defects using the low-frequency image. After NSST decomposition, the low-frequency component has less noise content, but its defect contour is relatively blurred. Use it to extract the main area information of the defect and reduce noise interference.

[0129] Take the low-frequency component L as the input of the pulse coupled neural network (PCNN), and use Shannon entropy to determine the optimal number of iterations T opt . After PCNN processing, obtain the coarsely segmented welding defect image D coarse .

[0130] Step3: Background suppression of the low-frequency image. Determine the threshold T according to the valley inflection point of the image histogram bg , Suppress the background part. Use a spline curve to fit the histogram to reduce the influence of local fluctuations on threshold selection. Welding defect images usually have low contrast and large background fluctuations, and the gray value of the weld bead part is generally large. The corresponding threshold can be found according to a valley inflection point of the image histogram curve, and the fluctuating background part is directly set to zero to reduce the interference of the background on defect extraction. In order to prevent misselection of the threshold caused by local fluctuations of the histogram curve, a spline curve is used to fit the histogram curve.

[0131] Step4: Construction of the high-frequency feature image. Use the NSST inverse transform for the background-suppressed low-frequency image L bg-inhand the high-frequency image H i are reconstructed to obtain the high-frequency feature image H feature .

[0132] Step5: Use the high-frequency feature image for defect fine segmentation. Perform threshold segmentation on H feature to obtain the initial segmentation result. Take the obtained image as the initial condition of the improved CV model and continuously approximate the optimal edge of the defect. The evolution equation of the level set function φ is as follows:

[0133]

[0134] ω(x) = {ω1, φ(x) > 0; ω2, φ(x) < 0};

[0135] In the formula, z represents the image gray value; c1 and c2 are the average gray values of the image in the internal and external regions of the curve C; μ, λ1, λ2 are weight parameters, δ is the Dirac function. φ is the level set function. ω1 raises the height of the 0-level set plane, causing the boundary to move towards the direction with a larger gray value, while ω2 is used to lower the height of the 0-level set plane, causing the boundary to move towards the direction with a smaller gray value. t is the time variable; represents the change rate of the level set function φ with respect to time t. That is to say, it describes how the level set evolves over time.

[0136] Step6: Fuse the results obtained from the high-frequency component segmentation. Perform a logical AND or OR operation on the defect image D extracted from the high-frequency feature image high and the defect image D extracted from the low-frequency image low to obtain the final defect image D final . Use the Sobel operator to extract the fine edges of the defects in D final . On the one hand, it can extract the fine edges of the defects, and on the other hand, it can effectively remove background noise.

[0137] In this technical solution, the feature extraction module can effectively extract the detailed information of welding defects from the welding image, including the approximate area, fine edges, and contours of the defects. This method combines the multi-scale and multi-directional decomposition ability of NSST and the edge optimization ability of the improved CV model, improving the accuracy and reliability of welding defect detection.

[0138] Furthermore, the infrared data processing module: processes and analyzes the detected data to identify the welding problems of the steel structure, including the following:

[0139] Data reception and preprocessing: The infrared welding transmittance detector first captures the infrared radiation data of the welding part, and this data contains the key information of welding quality.

[0140] The data processing module will receive these raw data and perform necessary preprocessing, such as removing noise, calibrating sensor responses, etc., to ensure the accuracy and reliability of the data.

[0141] Transmittance calculation and analysis: Transmittance refers to the proportion of light passing through the welded part and is an important indicator for evaluating welding quality. The infrared data processing module will calculate the transmittance of the welded part based on the received infrared radiation data.

[0142]

[0143] Among them, Tavg is the average transmittance of the welding area, Ti is the transmittance of the i-th detection point, which is the measured value of a single data point in the infrared welding transmittance detection. N is the total number of detection points, representing the number of points measured in the welding area.

[0144] By analyzing the transmittance data, it is possible to preliminarily determine whether the welding quality is qualified. The higher the transmittance, usually the better the welding quality.

[0145] Welding problem identification: The data processing module will conduct in-depth analysis on the transmittance data and, in combination with preset thresholds and algorithms, identify potential welding problems.

[0146] These welding problems may include welding defects (such as cracks, pores, slag inclusions, etc.), uneven welding, too high or too low welding temperature, etc. The identified welding problems will be recorded in detail and can be presented to the operator through the user interface or in the form of a report.

[0147] In this technical solution, by calculating the average value of the transmittance of all detection points within the welding area, it helps to evaluate the overall quality of the entire welding area, because the average transmittance can reflect the consistency and uniformity of the welded joint. If the average transmittance is lower than a preset threshold, it may indicate that there are quality problems with the welding and further inspection or repair is required.

[0148] Furthermore, the data information collected by the data acquisition module includes:

[0149] I. Data related to the welding of bridge steel structures

[0150] Welding position information: Precisely record the position of each welding point, including its specific coordinates on the bridge steel structure (such as coordinates on the CAD drawing) or actual geographical coordinates (such as longitude and latitude). Describe the steel structure layout and connection conditions near the welding point.

[0151] Weld bead thickness and quality data: Measure and record the actual thickness of the weld bead and compare it with the design requirements.

[0152] Evaluate the quality of the weld bead, including the flatness, continuity, uniformity, etc. of the weld bead.

[0153] Welding anomaly images: Collect and organize images of various welding anomalies, such as cracks, welding deformations, false welds, pores, arc pits, lack of fusion, etc.

[0154] Perform detailed annotation on each image, including information such as the anomaly type, location, size, etc.

[0155] Establish a reference sample database: Organize all the collected welding-related data (including drawings, technical requirements, images, etc.) to establish a unified database. The database should be convenient for querying and retrieval for subsequent analysis and processing.

[0156] II. Bridge structure and design data:

[0157] CAD drawings and 3D models: Collect the CAD drawings and 3D models of the bridge to understand the overall layout of the bridge, the positions and dimensions of key components. Ensure the accuracy and integrity of the drawings and models for subsequent simulation and path planning. Clearly define the purpose and scope of the inspection, and determine the accuracy requirements for the inspection, such as the resolution of weld inspection, measurement error, etc. According to the structural characteristics and design requirements of the bridge, determine the key areas and details that need to be focused on and inspected. Describe and record these areas in detail for subsequent UAV cruising and data analysis.

[0158] III. Bridge surrounding environment information:

[0159] Meteorological data: Collect meteorological data of the bridge surrounding environment, such as wind speed, wind direction, temperature, humidity, etc. These data will directly affect the flight stability and safety of the UAV and need to be monitored and recorded in real time.

[0160] Topographic and geomorphic data: Collect topographic and geomorphic data of the bridge surrounding area, such as terrain height, slope, geomorphic features, etc. These data will affect the cruising path planning and flight safety of the UAV and need to be measured and analyzed in detail.

[0161] Obstacle information: Record the obstacle information in the bridge surrounding environment, such as buildings, trees, wires, etc. These obstacles will affect the flight trajectory and obstacle avoidance strategy of the UAV and need to be accurately represented in the environmental model.

[0162] Furthermore, the route planning module: Based on the bridge structure and inspection requirements, use the A* algorithm to optimize the UAV cruising path, including the following steps:

[0163] 1. Recording of bridge numbers and geographical location information: Assign unique numbers to different bridges for subsequent management and identification. Record geographical location information such as the longitude, latitude, and height of each bridge, which helps with UAV positioning and navigation.

[0164] 2. Bridge structure analysis: Collect structural data such as CAD drawings and 3D models of the bridge to understand the overall layout of the bridge, the positions and dimensions of key components. Clearly define the purpose, scope, and accuracy requirements of the inspection, and determine the key areas and details of the bridge to be inspected.

[0165] 3. UAV cruise path planning:

[0166] Environmental modeling: Based on the geographical location information and structural data of the bridge, establish an environmental model, including the bridge structure, the surrounding environment (such as buildings, trees, power lines, etc.), and flight restricted areas (such as no-fly zones, restricted heights, etc.).

[0167] Path search: Use the A* algorithm to search for paths in the environmental model. The A* algorithm selects the next moving direction by comprehensively considering the actual cost and heuristic estimated cost of the path to achieve the goal of finding the optimal path. Set the starting point (UAV takeoff point), the ending point (bridge inspection point), and obstacles (bridge structure, obstacles in the surrounding environment). Design a cost function according to inspection requirements and environmental factors, such as flight time, flight distance, obstacle avoidance cost, etc. The A* algorithm will continuously search and update the path from the starting point to the ending point until the optimal path is found or it is determined that there is no feasible path.

[0168] Path optimization: Smooth the optimal path found by the A* algorithm to reduce unnecessary turns and undulations, and improve the flight stability and safety.

[0169] 4. Cruise route verification and adjustment:

[0170] Simulation verification: Use simulation software to simulate the UAV cruising according to the planned path to verify the feasibility and effectiveness of the path. According to the results of the simulation verification, make necessary adjustments to the path to adapt to different bridge structures and inspection requirements.

[0171] Pre-flight preparation: According to the planned cruise route, make pre-flight preparations for the UAV, including checking the equipment status, setting flight parameters, etc. During the UAV cruise, through the sensors on the UAV and the ground control system, monitor the position, speed, attitude and other parameters of the UAV in real time to ensure that the UAV flies stably according to the planned path.

[0172] Furthermore, for environmental modeling, remote sensing technologies such as Light Detection and Ranging (LiDAR) and unmanned aerial vehicle (UAV) aerial photography are used to obtain three-dimensional data of the bridge and its surrounding environment. Point cloud processing software (such as PCL, CloudCompare, etc.) is used to process the acquired three-dimensional data to extract the structural information of the bridge and the obstacle information of the surrounding environment. The processed data is imported into Blender modeling software to automatically generate an environmental model.

[0173] Specifically, it includes the following steps:

[0174] 1. Data collection: Use LiDAR equipment to scan the bridge and its surrounding environment from different angles to obtain high-precision three-dimensional point cloud data. Deploy UAVs equipped with high-definition cameras and / or LiDAR sensors to conduct aerial photography of the bridge to collect images and three-dimensional data.

[0175] 2. Point cloud preprocessing: Synchronize the LiDAR point cloud data and the UAV aerial photography data in time and register them spatially. Use point cloud processing software (such as PCL, CloudCompare) to remove noise and outliers in the point cloud.

[0176] 3. Structural information extraction: Separate the bridge structure points from the background points according to the attributes of the point cloud (such as reflection intensity, height, density). Identify the key structural features of the bridge, such as beams, columns, joints, etc., through feature extraction.

[0177] 4. Obstacle detection: Analyze obstacles in the surrounding environment, such as buildings, trees, terrain, etc. Mark the obstacles that may affect bridge detection and maintenance in the point cloud data.

[0178] 5. Data processing and modeling: Optimize the extracted structural information and obstacle information to remove redundant data. Import the processed data into modeling software such as Blender to automatically generate a three-dimensional model of the bridge and its surrounding environment.

[0179] 6. Model refinement: Add details of the bridge structure, such as textures, materials, etc., to the three-dimensional model. Simulate the impact of environmental factors on the bridge, such as lighting, shadows, etc.

[0180] 7. Model verification and application: Verify the accuracy and reliability of the model by comparing it with the actual bridge. Use the generated environmental model for bridge structural analysis, health monitoring, maintenance planning, etc.

[0181] Furthermore, the Euclidean distance and cosine similarity used by the anomaly recognition module in feature matching are mathematical tools for measuring the difference between two vectors. Combine the formulas with welding problems such as cracks, welding deformations, lack of fusion, porosity, craters, and incomplete fusion to identify welding quality problems, according to the following steps:

[0182] Feature Vectorization: First, convert the welding defects (such as cracks, deformations, pores, craters, lack of fusion, etc.) identified in the image into numerical feature vectors. These features may include the size, shape, texture, location, etc. of the defects.

[0183] Size Features: Measure the size features such as the width and length of defects like cracks and pores.

[0184] Shape Features: Extract the shape descriptors of the defects, such as roundness, rectangularity, etc.

[0185] Texture Features: Analyze the texture features of the defect area, such as roughness, uniformity, etc.

[0186] Location Features: Record the coordinate positions of the defects within the welding area.

[0187] Establish a Reference Sample Library: Create an image database containing normal welds and various known welding defects, and extract feature vectors for these images as reference samples. Collect images of normal welds and their feature vectors. Collect defect samples, including welding images with defects such as cracks, welding deformations, false soldering, pores, craters, lack of fusion, etc. and their feature vectors.

[0188] Feature Matching Process: For a new welding image, extract similar feature vectors, and then use the following formula to calculate the difference between the feature vectors of the new image and the reference samples:

[0189]

[0190] Cosine Similarity = (A·B) / (||A||||B||);

[0191] In the formula, d(A,B) is the Euclidean distance between vectors A and B. A and B are two feature vectors representing the features in the welding image. Ai and Bi are the values of the i-th feature in vectors A and B respectively. n is the total number of features in feature vectors A and B. Cosine Similarity is the cosine similarity between vectors A and B, representing the cosine value of the angle between them. A·B is the dot product of vectors A and B. ||A|| and ||B|| are the norms (usually the Euclidean norm) of vectors A and B, that is, the lengths of the vectors.

[0192] The Euclidean distance is used to measure the straight-line distance between feature vectors. The smaller the distance, the more similar. The cosine similarity measures the angle between feature vectors. The closer the cosine value is to 1, the more similar.

[0193] Calculations using formulas help us quantify the differences between feature vectors, thereby identifying welding defects. Feature vectors A and B may contain various features extracted from welding images, such as the width and length of cracks, the angle of welding deformation, the texture of the soldering void area, etc. By comparing these feature vectors with those of reference samples (welding images of normal welding or known defects), we can evaluate the welding quality and identify potential welding problems.

[0194] Threshold setting: Set a threshold to distinguish whether the welding quality is qualified. If the calculated distance or similarity exceeds this threshold, it is considered that there is a problem with the welding.

[0195] Set a distance threshold Td, and if it is exceeded, it is considered that there is a welding defect.

[0196] Set a similarity threshold Tc, and if it is lower than this threshold, the welding quality is considered unqualified.

[0197] Defect identification: Based on the results of feature matching, identify the types of defects in the welding image. For example, if the similarity between a certain feature vector and the feature vector of a crack in the database is very low, it is considered that there is no crack in this welding.

[0198] Size and position location: For the identified welding defects, use image processing techniques to determine their specific positions and sizes in the image, providing information for subsequent repair or further analysis.

[0199] Boundary detection: Use image processing techniques and edge detection algorithms such as Canny, Sobel or others to identify the boundaries of defects in the image. Convert the image into a binary image to facilitate clear identification of the boundaries. Implement an automatic boundary tracking algorithm to determine the defect contour.

[0200] Size measurement: Calculate the size of the defect, such as length, width, etc.

[0201] Defect area calculation: Calculate the number of pixels in the binarized defect area and convert it into the actual area. Adefect = Pixel Count × Pixel Area where Adefect is the defect area, Pixel Count is the number of pixels in the defect area, and Pixel Area is the actual area corresponding to each pixel.

[0202] Length and width measurement: For defects with regular shapes, directly measure their length and width; for irregular shapes, use methods such as the minimum bounding rectangle or fitted ellipse. <s

[0203] L = max(x) - min(x);

[0204] W = max(y) - min(y);

[0205] L and W are the length and width of the defect.

[0206] Position determination: Record the specific coordinate position of the defect in the welding image.

[0207] Furthermore, the data fusion module combines the welding anomalies identified by the anomaly recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure, including the following steps;

[0208] Data preprocessing: Clean the data to remove invalid or incorrect data points. Convert data in different formats into a unified format for easy processing. Normalize the data to eliminate the influence of different dimensions.

[0209] Data fusion processing:

[0210] Feature-level fusion: F combined = w1F visual + w2F IR ; where F combined is the fused feature vector, F visual and F IR are the feature vectors from the anomaly recognition module and the infrared data processing module respectively, and w1 and w2 are the corresponding weights.

[0211] Welding quality problem identification: Quality Issue = {[Yes, if F combined <T or D = Reject]; [No, otherwise]}, where Quality Issue is a boolean variable indicating whether there is a welding quality problem, T is a preset threshold, and Reject is the result of decision-level fusion.

[0212] Classification and quantification: C = h(Q); where C is the classification result of the welding quality problem, h is the classification function, and the problem is divided into different categories and severities according to the value of Q.

[0213] Result output: Output the evaluation results of the data fusion module, including the overall evaluation of the welding quality, the detailed information of the problem area, and the recommended repair measures.

[0214] In this technical solution, the data fusion module can integrate the results of two independent modules to provide a more comprehensive and accurate welding quality assessment. This method can improve the reliability and efficiency of the welding quality monitoring system and help operators quickly identify and handle potential welding problems.

[0215] Referring to Figure 2 , an angular position adjustment mechanism is provided on the drone 1;

[0216] The angular position adjustment mechanism includes an electric telescopic rod 2, an electric turntable 3, and a motor 4;

[0217] An electric telescopic rod 2 is fixedly arranged on the unmanned aerial vehicle 1; an electric turntable 3 is fixedly arranged on the movable rod of the electric telescopic rod 2;

[0218] The output end of the electric turntable 3 is fixedly provided with a motor 4, the output end of the motor 4 is coaxially and fixedly provided with a shaft rod 9, and a high-definition camera 5 is fixedly arranged on the shaft rod 9;

[0219] In this technical solution, by starting the electric telescopic rod 2 to drive the electric turntable 3, the motor 4, and the high-definition camera 5 to move, the position where the high-definition camera 5 extends is adjusted;

[0220] By driving the motor 4 and the high-definition camera 5 to rotate through the electric turntable 3, and driving the high-definition camera 5 to rotate in another direction through the motor 4, the angle of the high-definition camera 5 is adjusted, so that the high-definition camera 5 is aligned with the welding position, facilitating the acquisition of high-definition images. Especially when detecting the welding at the corner or the upper end of the top, adjusting the angle of the high-definition camera 5 can acquire high-definition images.

[0221] A welding defect marking mechanism is arranged on the unmanned aerial vehicle 1, and the welding defect marking mechanism includes a welding defect marking liquid storage tank 7, a suction pump 8, and a spray pipe 10;

[0222] A welding defect marking liquid storage tank 7 is fixedly arranged on the unmanned aerial vehicle 1, and a suction pump 8 is fixedly arranged on the welding defect marking liquid storage tank 7; the input end of the suction pump 8 is communicated with the inside of the welding defect marking liquid storage tank 7 through a pipeline; the welding defect marking liquid storage tank 7 is filled with red welding defect marking liquid, and the welding defect marking liquid can be selected as red or yellow according to needs. The welding defect marking liquid can also be red or yellow spray paint.

[0223] A spray pipe 10 is fixedly arranged on the shaft rod 9; the spray pipe 10 is communicated with the output end of the suction pump 8 through a hose;

[0224] In this technical solution, when it is detected that there are welding quality defects on the bridge steel structure, the welding defect marking liquid in the welding defect marking liquid storage tank 7 is sucked into the spray pipe 10 through the suction pump 8, and then sprayed near the welding defect, facilitating quickly finding the position of the welding defect during later maintenance.

[0225] The present invention provides a method for detecting the welding quality of bridge steel structures, including the following steps:

[0226] S1. The data acquisition module collects a large amount of relevant data (including drawings and welding requirements) of the welding of bridge steel structures, including information such as the welding position, welding method, and weld bead thickness; includes images of normal welding and welding with abnormal conditions, and annotates the images as reference samples;

[0227] S2. The route planning module optimizes the UAV cruise path using an algorithm based on the bridge structure and inspection requirements to plan the UAV cruise route.

[0228] S3. The inspection module conducts cruise inspections on the welding quality of the steel structure of the bridge through the UAV.

[0229] S4. The image preprocessing module preprocesses the collected images, including filtering and denoising, grayscale conversion, image segmentation, and image enhancement. The feature extraction module extracts features from the preprocessed images, extracting features related to welding anomalies, including welding problems such as cracks, welding deformation, lack of fusion, porosity, craters, and incomplete fusion.

[0230] S5. The anomaly recognition module compares and analyzes the images after feature extraction with reference samples to identify welding anomalies, including welding problems such as cracks, welding deformation, lack of fusion, porosity, craters, and incomplete fusion; and identifies the size and location of the welding anomalies.

[0231] S6. The infrared data processing module processes and analyzes the data detected by the infrared welding light transmittance detector to identify welding problems in the steel structure.

[0232] S7. The data fusion module combines the welding anomalies identified by the anomaly recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify welding quality problems in the steel structure.

[0233] S8. When welding quality problems are detected, the alarm module issues an alarm and records the location information of the welding quality problems. By starting the suction pump 8, the welding defect marking liquid in the welding defect marking liquid storage tank 7 is sucked into the nozzle 10 and then sprayed near the welding defect, facilitating the quick location of the welding defect during later maintenance.

[0234] The working principle of a bridge steel structure welding quality detection system of the present invention is as follows: The data acquisition module collects a large amount of relevant data on the welding of bridge steel structures (including drawings and welding requirements), including information such as the welding position, welding method, and weld bead thickness; includes images of normal welding and welding with abnormal conditions, and annotates the images as reference samples; the route planning module uses algorithms to optimize the UAV cruising path based on the bridge structure and detection requirements, and plans the cruising route of the UAV; the inspection module conducts cruising detection on the welding quality of the bridge steel structure through the UAV; the image preprocessing module preprocesses the collected images, including filtering and denoising, grayscale conversion, image segmentation, and image enhancement. The feature extraction module extracts features from the preprocessed images, extracting features related to welding abnormal conditions, including welding problems such as cracks, welding deformation, lack of fusion, pores, craters, and incomplete fusion; the abnormal recognition module compares and analyzes the images after feature extraction with the reference samples to identify welding abnormal conditions, including welding problems such as cracks, welding deformation, lack of fusion, pores, craters, and incomplete fusion; identifies the size and position of the welding abnormal conditions; the infrared data processing module processes and analyzes the data detected by the infrared welding light transmittance detector to identify the welding problems of the steel structure; the data fusion module combines the welding abnormal conditions identified by the abnormal recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure; when welding quality problems are detected, the alarm module issues an alarm and records the position information of the welding quality problems. By starting the suction pump 8, the welding defect marking liquid in the welding defect marking liquid storage tank 7 is sucked into the spray pipe 10 and then sprayed near the welding defect, facilitating the quick location of the welding defect during later maintenance.

[0235] The present invention uses UAVs for automated inspection, reducing the need for manual inspection, increasing the inspection speed, and improving work efficiency. By combining the visual inspection and infrared inspection results through the data fusion module, the accuracy of identifying welding quality problems is improved. By using the welding defect marking liquid, this method can visually mark the defect position on the physical structure, facilitating subsequent maintenance. The precise positioning and marking of defects reduce the time for maintenance personnel to search for and identify problems, improving the maintenance efficiency. The bridge steel structure welding quality detection method not only improves the accuracy and efficiency of detection but also provides strong technical support for the maintenance and management of bridges. The extension length and angle of the high-definition camera 5 are adjusted through the angle position adjustment mechanism, aligning the high-definition camera 5 with the welding position to facilitate the acquisition of high-definition images. Especially when detecting the welding at the corners or the upper end of the top, adjusting the angle of the high-definition camera 5 can acquire high-definition images, expanding the scope of use.

[0236] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than limiting it; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting the welding quality of bridge steel structures, characterized in that, Including the following steps: S1. The data acquisition module collects a large amount of data related to the welding of bridge steel structures, including welding positions, welding methods, and bead thickness information; Including images of normal welding and welding with abnormal conditions, and annotating the images as reference samples; S2. The route planning module optimizes the UAV cruise path using an algorithm based on the bridge structure and inspection requirements, and plans the cruise route of the UAV; S3. The inspection module conducts cruise inspections on the welding quality of the bridge steel structure through the UAV; S4. The image preprocessing module preprocesses the collected images, and the feature extraction module extracts features from the preprocessed images, extracting features related to welding abnormal conditions; S5. The abnormal recognition module compares and analyzes the images after feature extraction with the reference samples to identify welding abnormal conditions; identifies the size and position of welding abnormal conditions; S6. The infrared data processing module processes and analyzes the data detected by the infrared welding light transmittance detector to identify welding problems of the steel structure; S7. The data fusion module combines the welding abnormal conditions identified by the abnormal recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure; S8. When welding quality problems are detected, the alarm module issues an alarm and records the location information of the welding quality problems; Spray red marks near the welding defects through the welding defect marking mechanism to facilitate quickly finding the welding defect location during later maintenance.

2. The method for detecting the welding quality of a bridge steel structure according to claim 1, characterized in that: Step S4 includes the following steps: S41. Perform single-layer non-subsampled Shearlet decomposition on the welding image I to obtain a low-frequency component L and six high-frequency components H i , perform non-subsampled Laplacian pyramid decomposition using the "Maxflat" filter, and perform direction analysis using the "Meyer" window function; S42. Coarsely segment the defects using the low-frequency image, extract the information of the main defect regions, and reduce noise interference; use the low-frequency component L as the input of the Pulse Coupled Neural Network (PCNN), and determine the optimal number of iterations T using Shannon entropy opt ; After being processed by PCNN, obtain the image D of the welded defects after coarse segmentation coarse ; S43. Determine the threshold T based on the valley inflection point of the image histogram bg , and suppress the background part; use a spline curve to fit the histogram to reduce the influence of local fluctuations on the threshold selection; S44. High-frequency feature image construction: Use the NSST inverse transform to reconstruct the low-frequency image L bg-inh and the high-frequency image H i after background suppression to obtain the high-frequency feature image H feature ; S45. Use the high-frequency feature image for defect fine segmentation, and perform threshold segmentation on H feature to obtain the initial segmentation result, and use the obtained image as the initial condition of the improved CV model to continuously approximate the optimal edge of the defect; S46. Combine the results obtained by segmenting the high-frequency components, and extract the defective image D from the high-frequency feature image high and the defective image D extracted from the low-frequency image low Perform a logical AND or OR operation to obtain the final defective image D final ; Use the Sobel operator to extract the fine edges of the defects in D final ​ 3. The method for detecting the welding quality of bridge steel structures according to claim 2, characterized in that: In step S45, the evolution equation of the level set function φ is as follows: ω(x) = {ω1, φ(x) > 0; ω2, φ(x) < 0}; Wherein, z represents the image gray value; c1 and c2 are the average gray values of the images in the inner and outer regions of curve C; μ, λ1, and λ2 are weight parameters, δ is the Dirac function; φ is the level set function; ω1 increases the height of the 0-level set plane, causing the boundary to move in the direction of the larger gray value, and ω2 is used to reduce the height of the 0-level set plane, causing the boundary to move in the direction of the smaller gray value; t is the time variable; represents the change rate of the level set function φ with respect to time t.

4. The method for detecting the welding quality of bridge steel structures according to claim 3, characterized in that: Step S2 includes the following steps: S21. Bridge number and geographical location information recording: Assign unique numbers to different bridges for subsequent management and identification, and record the longitude, latitude, and height information of each bridge; S22. Bridge structure analysis: Collect CAD drawings and three-dimensional model structure data of the bridge to understand the overall layout, positions and dimensions of key components of the bridge; clarify the purpose, scope, and accuracy requirements of the inspection, and determine the key areas and details of the bridge to be inspected; S23. UAV cruise path planning: Based on the geographical location information and structure data of the bridge, establish an environmental model, including the bridge structure and the surrounding environment; use the A* algorithm to search for paths in the environmental model; adopt the A* algorithm to continuously search for and update the path from the starting point to the ending point until the optimal path is found or it is determined that there is no feasible path; S24. Cruise route verification and adjustment: Use simulation software to simulate the UAV cruising according to the planned path to verify the feasibility and effectiveness of the path; according to the results of the simulation verification, make necessary adjustments to the path to adapt to different bridge structures and inspection requirements.

5. The method for detecting the welding quality of a bridge steel structure according to claim 3, characterized in that: Step S5 includes the following steps: S51. Feature vectorization: Convert the welding defects identified in the image into numerical feature vectors; the features include the size, shape, texture, and position of the defects; S52. Establish a reference sample library: Create an image database containing normal welds and various known welding defects, and extract feature vectors for these images as reference samples; S53. Feature matching process: For a new welding image, extract similar feature vectors, and then calculate the difference between the feature vectors of the new image and the reference samples; S54. Threshold setting: Set a threshold to distinguish whether the welding quality is qualified; if the calculated distance or similarity exceeds this threshold, it is considered that there is a problem with the weld; S55. Defect identification: Identify the type, size, etc. of the defects in the welding image according to the results of feature matching.

6. The method for detecting the welding quality of bridge steel structures according to claim 5, characterized in that: In step S53, the following formula is used to calculate the difference between the new image feature vector and the reference sample: Cosine Similarity = (A·B) / (||A||||B||); Where, d(A,B) is the Euclidean distance between vectors A and B; A and B are two feature vectors representing features in the welding image; A i and B i are the values of the i-th feature in vectors A and B respectively; n is the total number of features in feature vectors A and B; Cosine Similarity is the cosine similarity between vectors A and B, representing the cosine value of the angle between them; A·B is the dot product of vectors A and B; ||A|| and ||B|| are the norms of vectors A and B respectively; Euclidean distance is used to measure the straight-line distance between feature vectors, and the smaller the distance, the more similar; cosine similarity measures the angle between feature vectors, and the closer the cosine value is to 1, the more similar.

7. The method for detecting the welding quality of bridge steel structures according to claim 2, characterized in that: Step S7 includes the following steps: S71. Data preprocessing: Clean the data to remove invalid or incorrect data points; convert data in different formats into a unified format, and perform normalization processing on the data to eliminate the influence of different dimensions; S72. Data fusion processing: Includes feature-level fusion and decision-level fusion; Feature-level fusion: F combined = w1F visual + w2F IR ; where F combined is the fused feature vector, F visual and F IR are the feature vectors from the anomaly recognition module and the infrared data processing module respectively, and w1 and w2 are the corresponding weights; S73. Welding quality problem identification: Quality Issue = {[Yes, if Fcombined < T or D = Reject]; [No, otherwise]}, where Quality Issue is a boolean variable indicating whether there is a welding quality problem, T is a preset threshold, and Reject is the result of decision-level fusion; S74. Result output: Output the evaluation results of the data fusion module, including the overall evaluation of welding quality, detailed information on the problem area, and recommended repair measures.

8. The method for detecting the welding quality of bridge steel structures according to claim 1, characterized in that: An angle position adjustment mechanism is provided on the drone; The angle position adjustment mechanism includes an electric telescopic rod, an electric turntable, and a motor; An electric telescopic rod is fixedly provided on the drone; an electric turntable is fixedly provided on the movable rod of the electric telescopic rod; The output end of the electric turntable is fixedly provided with a motor, the output end of the motor is coaxially fixedly provided with a shaft rod, and a high-definition camera is fixedly provided on the shaft rod.

9. The method for detecting the welding quality of a bridge steel structure according to claim 8, characterized in that: A welding defect marking mechanism is provided on the drone, and the welding defect marking mechanism includes a welding defect marking liquid storage tank, a suction pump, and a spray pipe; A welding defect marking liquid storage tank is fixedly provided on the drone, and a suction pump is fixedly provided on the welding defect marking liquid storage tank; the input end of the suction pump is communicated with the inside of the welding defect marking liquid storage tank through a pipeline; the welding defect marking liquid storage tank is filled with red welding defect marking liquid; A spray pipe is fixedly provided on the shaft rod; the spray pipe is communicated with the output end of the suction pump through a hose.

10. A bridge steel structure welding quality detection system, comprising: A data acquisition module, a route planning module, an inspection module, an image preprocessing module, a feature extraction module, an anomaly recognition module, a data fusion module, and an alarm module; characterized in that: The data acquisition module: collects a large amount of data related to the welding of bridge steel structures, including the welding position, welding method, and bead thickness information; includes images of normal welding and welding with abnormal conditions, and annotates the images to establish a reference sample database; The route planning module: based on the bridge structure and inspection requirements, uses an algorithm to optimize the UAV cruise path and plan the UAV cruise route; The inspection module: includes a UAV, on which a high-definition camera, a GPS locator, and an infrared welding light transmittance detector are set; conducts cruise inspection on the welding quality of the bridge steel structure; The image preprocessing module: preprocesses the collected images, including filtering and denoising, grayscale conversion, image segmentation, and image enhancement; The feature extraction module: extracts features from the preprocessed images, and extracts features related to welding abnormal conditions, including cracks, welding deformation, lack of fusion, porosity, craters, and incomplete fusion problems; The anomaly recognition module: compares and analyzes the images after feature extraction with the reference samples to identify welding abnormal conditions, including cracks, welding deformation, lack of fusion, porosity, craters, and incomplete fusion problems; identifies the size and position of the welding abnormal conditions; The infrared data processing module: processes and analyzes the data detected by the infrared welding light transmittance detector to identify the welding problems of the steel structure; The data fusion module: combines the welding abnormal conditions identified by the anomaly recognition module with the welding quality results detected by the infrared data processing module to quickly and accurately identify the welding quality problems of the steel structure; The alarm module: includes an alarm. When welding quality problems are detected, an alarm is issued, and the position information of the welding quality problems is recorded; The PLC control module is network-connected to the route planning module, the inspection module, the data fusion module, and the alarm module, and is network-connected to the remote control center through a wireless signal.

Images