Integrated assembly type steel structure assembly quality detection method based on BIM

Abstract

The invention provides an integrated assembly type steel structure assembly quality detection method based on BIM, and relates to the technical field of steel structure assembly, and the method comprises the steps: obtaining a multi-source data set; a bidirectional region fusion image is obtained, a weld feature set is determined, and an image quality detection result is obtained; constructing a three-dimensional region contour, generating a region three-dimensional model, and generating a point cloud quality detection result; obtaining a gray level enhanced image, and determining a welding seam feature set; constructing a pixel feature matrix, inputting the pixel feature matrix into a preset template matrix library for similar traversal comparison, and determining an infrared quality detection result; and outputting a welding quality detection result of the target welding area. According to the method and the device, the technical problem of relatively low accuracy and efficiency of welding quality evaluation of the connection point due to the fact that quality detection is carried out mostly through a single detection method during steel structure assembly quality detection in the prior art can be solved, and the technical target of multi-source defect detection is achieved; the technical effect of improving the accuracy and efficiency of the welding quality evaluation of the connection points is achieved.

Description

Technical Field

[0001] This application relates to the field of steel structure assembly technology, and in particular to a BIM-based integrated prefabricated steel structure assembly quality inspection method. Background Technology

[0002] Steel structure assembly quality inspection is a key step in ensuring the safety, reliability, and durability of steel structure buildings. It involves inspecting each component of the steel structure to ensure that it meets design and construction standards during manufacturing and installation.

[0003] Currently, existing methods for inspecting the assembly quality of steel structures, given their complexity, mostly rely on single defect detection methods. This results in an inability to quickly and accurately inspect the welding quality of steel structure connection points, failing to meet expected inspection requirements. Therefore, a method is needed to address these issues.

[0004] In summary, existing technologies suffer from the problem that the quality inspection of steel structure assembly is mostly carried out using a single inspection method, resulting in low accuracy and efficiency in the assessment of welding quality at connection points, which further affects the quality of welding quality assessment. Summary of the Invention

[0005] The purpose of this application is to provide a BIM-based integrated prefabricated steel structure assembly quality inspection method to solve the technical problem that in the prior art, the quality inspection of steel structure assembly is mostly carried out by a single inspection method, resulting in low accuracy and efficiency of welding quality assessment at connection points, which further affects the quality of welding quality assessment.

[0006] In view of the above problems, this application provides a BIM-based integrated prefabricated steel structure assembly quality inspection method.

[0007] This application provides a BIM-based integrated prefabricated steel structure assembly quality inspection method, wherein the method includes: acquiring data from the target welding area at the connection point based on a multi-source device array to obtain a multi-source dataset, wherein the multi-source dataset includes a region image set, a point cloud dataset, and an infrared thermal imaging image; fusing the region image set according to a preset image fusion strategy to obtain a bidirectional region fused image; extracting features from the bidirectional region fused image through a feature recognition channel to determine the weld feature set; performing welding quality assessment based on the weld feature set to obtain an image quality inspection result; and extracting the contour from the bidirectional region fused image using a contour extraction model to construct a three-dimensional region contour, and performing welding quality inspection based on the three-dimensional region contour of the point cloud dataset. Point cloud fitting is performed on the data set to generate a 3D model of the region. Mesh defect identification is performed on the 3D model of the region to generate point cloud quality detection results. The infrared thermal imaging image is preprocessed according to a preset image processing strategy to obtain a grayscale enhanced image. The grayscale enhanced image is identified to determine the weld feature set. The preset image processing strategy includes grayscale processing, histogram equalization enhancement, and binarization processing. A pixel feature matrix is ​​constructed based on the weld feature set. The pixel feature matrix is ​​input into a preset template matrix library for similarity traversal comparison. The infrared quality detection results are determined based on the similarity comparison results. The image quality detection results, the point cloud quality detection results, and the infrared quality detection results are fused to output the welding quality detection results of the target welding area.

[0008] One or more technical solutions provided in this application have at least the following technical effects or advantages: Data is collected from the target welding area at the connection point using a multi-source device array to obtain a multi-source dataset, which includes a regional image set, a point cloud dataset, and an infrared thermal imaging image. The regional image set is fused according to a preset image fusion strategy to obtain a bidirectional regional fusion image. Features are extracted from the bidirectional regional fusion image using a feature recognition channel to determine the weld feature set. Welding quality is assessed based on the weld feature set to obtain image quality detection results. A contour extraction model is used to extract contours from the bidirectional regional fusion image to construct a three-dimensional regional contour. Point cloud fitting is performed on the point cloud dataset based on the three-dimensional regional contour to generate a three-dimensional regional model. Mesh defect identification is performed on the three-dimensional regional model to generate... Point cloud quality inspection results; the infrared thermal imaging image is preprocessed according to a preset image processing strategy to obtain a grayscale enhanced image, the grayscale enhanced image is identified, and the weld feature set is determined. The preset image processing strategy includes grayscale processing, histogram equalization enhancement, and binarization processing. A pixel feature matrix is ​​constructed based on the weld feature set, and the pixel feature matrix is ​​input into a preset template matrix library for similarity traversal comparison. The infrared quality inspection results are determined based on the similarity comparison results. The image quality inspection results, the point cloud quality inspection results, and the infrared quality inspection results are fused to output the welding quality inspection results of the target welding area, thereby achieving the technical goal of multi-source defect detection and improving the accuracy and efficiency of weld quality assessment at connection points.

[0009] The above description is merely an overview of the technical solution of this application. To better understand the technical means of this application and to facilitate its implementation according to the description, and to make the above and other objects, features, and advantages of this application more apparent, specific embodiments of this application are described below. It should be understood that the content described in this section is not intended to identify key or important features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent through the following description. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are merely exemplary. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating the BIM-based integrated prefabricated steel structure assembly quality inspection method of this application. Figure 2This is a flowchart illustrating the process of obtaining bidirectional region fusion images in the BIM-based integrated prefabricated steel structure assembly quality inspection method of this application. Detailed Implementation

[0012] This application provides a BIM-based integrated prefabricated steel structure assembly quality inspection method, which solves the technical problem in existing technologies where quality inspection of steel structure assembly mostly relies on a single inspection method, resulting in low accuracy and efficiency in weld quality assessment at connection points, further affecting the overall quality of weld quality assessment. It achieves the technical goal of multi-source defect detection, thereby improving the accuracy and efficiency of weld quality assessment at connection points.

[0013] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. It should be understood that this application is not limited to the exemplary embodiments described herein. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application. It should also be noted that, for ease of description, only the parts related to this application are shown in the accompanying drawings, not all of them.

[0014] Please see the appendix Figure 1 This application provides a BIM-based method for inspecting the assembly quality of integrated prefabricated steel structures, wherein the method specifically includes the following steps: Step 1: Based on the multi-source device array, data is collected from the target welding area at the connection point to obtain a multi-source dataset, which includes a regional image set, a point cloud dataset, and an infrared thermal imaging image.

[0015] Specifically, the target welding area is the region to be welded, and image acquisition begins as soon as welding is completed. Image acquisition devices such as cameras are used to obtain a set of regional images of the welding area, which are then used to inspect the appearance quality of the weld, such as the continuity, uniformity, and presence of defects. Three-dimensional coordinate data of the target welding area is acquired using devices such as laser scanners, resulting in a point cloud dataset. This provides geometric information about the welding area, helping to detect dimensional accuracy and structural consistency. Infrared thermal imaging cameras are used to acquire thermal distribution images of the welding area, revealing the thermal stress distribution during the welding process and assessing the internal quality and heat-affected zone of the weld. The regional image set, point cloud dataset, and infrared thermal imaging images are then integrated to obtain a multi-source dataset.

[0016] Step 2: The region image set is fused according to the preset image fusion strategy to obtain a bidirectional region fused image. Features are extracted from the bidirectional region fused image through the feature recognition channel to determine the weld feature set. Welding quality is evaluated based on the weld feature set to obtain the image quality detection result.

[0017] Specifically, based on a preset image fusion strategy, the frontal and lateral image sets are fused to obtain a bidirectional fused image. Combining image information from different perspectives provides a more comprehensive view of the weld. Assuming the welding area is a circular pipe, the fused image includes both frontal and lateral views. Feature extraction is performed on the bidirectional fused image through a feature recognition channel to identify and extract specific features of the weld, such as its location, shape, size, texture, and continuity, forming a weld feature set. Based on the extracted weld feature set, welding quality is evaluated using an expert system, machine learning model, or deep learning algorithm. This includes detecting weld defects such as cracks, porosity, and slag inclusions, as well as assessing the accuracy of weld geometry and dimensions. The resulting image quality detection result indicates whether the welding quality meets the predetermined standards.

[0018] Step 3: Use the contour extraction model to extract the contour of the bidirectional region fusion image, construct a three-dimensional region contour, perform point cloud fitting on the point cloud dataset based on the three-dimensional region contour, generate a three-dimensional region model, identify mesh defects in the three-dimensional region model, and generate point cloud quality detection results.

[0019] Specifically, a trained contour extraction model is used to extract contours from the bidirectional region fusion image, identifying the contours of the welding area in the image and constructing a 3D region contour based on forward and lateral fusion. Based on the extracted 3D region contour, a point cloud dataset is fitted to find the point cloud data that best matches the contour, generating a 3D model of the region, improving the accuracy of the 3D model and making it closer to the actual structure of the welding area. The generated 3D region model is then used to identify mesh defects such as cracks and porosity, specifically the volume ratio of cracks and porosity. Based on the identified crack and porosity feature sets, point cloud quality inspection results are generated, identifying potential welding defects.

[0020] Step 4: Preprocess the infrared thermal imaging image according to the preset image processing strategy to obtain a grayscale enhanced image, identify the grayscale enhanced image, and determine the weld feature set. The preset image processing strategy includes grayscale processing, histogram equalization enhancement, and binarization processing.

[0021] Specifically, a grayscale image has only one brightness value per pixel. The infrared thermal imaging image is converted to a grayscale image and processed. Uneven grayscale values ​​in the original image are adjusted to a histogram with a uniform grayscale distribution, improving the contrast and sharpness of the welding thermal image. By setting a threshold, pixels above the threshold are converted to white or black, and pixels below the threshold are converted to black or white, converting the grayscale image to a binary image where each pixel has only black or white, which helps to highlight weld features in the image. A grayscale-enhanced image is obtained by completing the preset image processing strategy through grayscale processing, histogram equalization enhancement, and binarization. In the preprocessed grayscale-enhanced image, image processing techniques such as edge detection, contour tracking, and morphological operations are used to identify weld beads, incomplete penetration, porosity, and undercut, determining the weld feature set, including attributes such as weld position, length, width, and shape.

[0022] Step 5: Construct a pixel feature matrix based on the weld feature set, input the pixel feature matrix into a preset template matrix library for similarity comparison, and determine the infrared quality detection result based on the similarity comparison result.

[0023] Specifically, a pixel feature matrix is ​​constructed based on the weld feature set, containing pixel-level feature information related to weld defects. The constructed pixel feature matrix is ​​then input into a pre-defined template matrix library for similarity comparison. This library includes different types of welding defects. Based on the similarity comparison results, the infrared quality detection results are determined, obtaining pixel features similar to the defect template, thereby inferring potential welding defects.

[0024] Step 6: Combine the image quality detection results, the point cloud quality detection results, and the infrared quality detection results to output the welding quality detection results of the target welding area.

[0025] Specifically, image quality inspection results, point cloud quality inspection results, and infrared quality inspection results are obtained through different detection technologies and methods, all of which include information about welding quality. These results are then fused, incorporating multi-source data, performing consistency checks, and resolving conflicts to ensure the accuracy and reliability of the final result. The welding quality inspection results for the target welding area are output, providing a comprehensive assessment of the welding quality.

[0026] The BIM-based integrated prefabricated steel structure assembly quality inspection method can achieve the technical goal of multi-source defect detection, and improve the accuracy and efficiency of welding quality assessment at connection points.

[0027] Furthermore, such as Figure 2 As shown, this application also includes: The region image set is divided based on the forward and lateral angles to obtain a forward region image set and a lateral region image set; a preset image fusion strategy is obtained, wherein the preset image fusion strategy includes a forward fusion strategy and a lateral fusion strategy; the forward fusion strategy is a mean strategy, and the lateral fusion strategy is a weighted calculation strategy for overlapping regions, wherein the weight of the overlapping region is inversely proportional to the image interference coefficient of the acquisition angle; the forward region image set and the lateral region image set are fused based on the forward fusion strategy and the lateral fusion strategy respectively to obtain a forward region fused image and a lateral region fused image, which are set as the bidirectional region fused image.

[0028] Specifically, a set of frontal angle images is acquired from the angle directly facing the target welding area, and a set of lateral area images is acquired from the side angle of the target welding area, and then the area image sets are divided.

[0029] Then, a preset image fusion strategy is obtained, which includes a forward fusion strategy and a lateral fusion strategy. The forward fusion strategy uses a mean averaging strategy, meaning the fused image is generated by averaging the pixel values ​​of all images from forward angles to reduce random noise and provide a smooth visual effect. The lateral fusion strategy uses a weighted calculation strategy for overlapping regions. Images from different lateral angles will have different levels of sharpness and interference in the overlapping regions. The weight of the overlapping region is inversely proportional to the interference coefficient of the image at the acquisition angle. This means that images with less interference have a higher weight during fusion. The weight of the overlapping region is determined by multiple interference factors, such as the light, reflection, and occlusion corresponding to the angle.

[0030] Next, a forward fusion strategy is used to fuse the forward region image set to generate a forward region fused image. A lateral fusion strategy is then used to fuse the lateral region image set to generate a lateral region fused image. Combining the forward and lateral region fused images forms a bidirectional region fused image, providing more comprehensive information, including details from the front and depth information from the sides, which helps to more accurately analyze and evaluate welding quality.

[0031] By combining image information from different angles, the accuracy and efficiency of weld area analysis are improved, serving as input for subsequent processing.

[0032] Furthermore, this application also includes: Using the attribute features of the target welding area as constraints, a set of sample forward images, a set of sample forward contours, a set of sample lateral images, and a set of sample lateral contours are obtained through network retrieval. The set of sample forward images and the set of sample forward contours are used as forward training data to perform supervised training and feedback verification on a convolutional neural network, resulting in a forward contour extraction unit that meets the expected convergence constraints. The set of sample lateral images and the set of sample lateral contours are used as lateral training data to perform supervised training and feedback verification on a convolutional neural network, resulting in a lateral contour extraction unit that meets the expected convergence constraints. The contour extraction model is constructed based on the forward contour extraction unit and the lateral contour extraction unit.

[0033] Specifically, based on the type, size, shape, and other attributes of the steel structure connection in the target welding area, sample data is retrieved via the internet, including a set of forward sample images, a set of forward sample contours, a set of lateral sample images, and a set of lateral sample contours, which are used to train a convolutional neural network.

[0034] Then, using the sample forward image set and sample forward contour set as forward training data, the convolutional neural network is trained under supervision to learn to predict the corresponding contours from the input images. The network predicts the contours and the actual sample contours for feedback verification to ensure that the output of the convolutional neural network meets the expected convergence constraints, thus obtaining a forward contour extraction unit that meets the expected convergence constraints.

[0035] Next, the sample lateral image set and sample lateral contour set are used as lateral training data to perform supervised training and feedback verification on the convolutional neural network, ensuring that the output of the convolutional neural network meets the expected convergence constraints.

[0036] Next, by combining the forward contour extraction unit and the lateral contour extraction unit, a contour extraction model is constructed, which can process images from different angles and extract the corresponding welding area contours.

[0037] By creating a model for extracting the contour of the welding area, it is possible to adapt to different perspectives and welding conditions, and use it in actual welding quality inspection to improve the accuracy and efficiency of inspection.

[0038] Furthermore, this application also includes: Using the three-dimensional region contour as a constraint, the point cloud dataset is randomly fitted to obtain a first point cloud fitting result, and a first fitting convergence of the first point cloud fitting result is calculated, wherein the first fitting convergence is the ratio of the number of point cloud data falling within the three-dimensional region contour to the total number of point cloud data in the point cloud dataset; again using the three-dimensional region contour as a constraint, the point cloud dataset is randomly fitted to obtain a second point cloud fitting result and a second fitting convergence; if the first fitting convergence is greater than the second fitting convergence, the first point cloud fitting result is taken as the current optimal point cloud fitting result, otherwise if the first fitting convergence is less than or equal to the second fitting convergence, the second point cloud fitting result is taken as the current optimal point cloud fitting result; iterative point cloud fitting is performed until a preset number of fittings is met, and the current optimal point cloud fitting result is output to construct the three-dimensional model of the region.

[0039] Specifically, a three-dimensional region contour is used as a constraint to randomly fit the point cloud dataset, resulting in the first point cloud fitting result. The first fitting convergence is calculated by comparing the ratio of the number of point cloud data points falling within the three-dimensional region contour to the total amount of data in the point cloud dataset. A higher ratio indicates higher fitting accuracy, and vice versa.

[0040] Then, the point cloud dataset is randomly fitted again using the 3D region contour as a constraint to obtain a second point cloud fitting result and a second fitting convergence. The second fitting convergence is calculated by comparing the ratio of the number of point cloud data points falling within the 3D region contour to the total amount of data in the point cloud dataset.

[0041] Next, the convergence of the first and second fittings is compared. If the convergence of the first fitting is greater than that of the second fitting, the first point cloud fitting result is selected as the current optimal point cloud fitting result; otherwise, the second point cloud fitting result is selected.

[0042] Next, the point cloud fitting is iterated repeatedly to find a better fit. This iteration continues until a preset number of fitting iterations is met. The preset number of fitting iterations is set based on specific accuracy requirements and computational resources. When the preset number of iterations is reached, the current optimal point cloud fitting result is output, obtaining the point cloud data that best matches the contour. A 3D model of the region is then constructed for subsequent analysis and evaluation, such as structural integrity checks and dimensional accuracy measurements.

[0043] By using the 3D region contour as a constraint to optimize the fitting of point cloud data, the accuracy of 3D model construction can be improved, and complex point cloud data can be processed effectively.

[0044] Furthermore, this application also includes: Obtain preset mesh defect indices, including crack indices and porosity indices; identify mesh defects in the three-dimensional model of the region based on the preset mesh defect indices to determine crack feature sets and porosity feature sets; evaluate the regional welding quality based on the crack feature sets and porosity feature sets to obtain the point cloud quality detection results.

[0045] Specifically, preset mesh defect indices are obtained as standards for evaluating weld quality, including crack indices and porosity indices. These indices are set according to industry standards or specific application requirements.

[0046] Then, using preset mesh defect indices, mesh defect identification is performed on the regional 3D model. Detailed mesh analysis is conducted on the 3D model to identify potential cracks and pores, determining crack and pore feature sets, including detailed information on the identified cracks and pores, such as location, size, and shape. Crack features include the number and size of cracks, while pore features include the number and size of pores.

[0047] Next, based on the crack feature set and porosity feature set, regional welding quality is assessed. According to preset indicators, the identified defects are quantitatively analyzed to determine the degree of influence of the crack feature set and porosity feature set on welding quality, and then point cloud quality inspection results are generated to determine whether the welding meets the quality standard or requires further repair or re-welding.

[0048] By identifying and assessing welding defects to ensure the safety and reliability of structures, quality problems in the welding process can be effectively detected by combining three-dimensional models and mesh defect identification technology.

[0049] Furthermore, this application also includes: The weld feature set includes at least weld bead features, incomplete penetration features, porosity features, and undercut features. A pixel feature matrix is ​​constructed based on the weld feature set, and the pixel feature matrix is ​​compared with multiple unqualified template matrices in a preset template matrix library to output multiple similarity scores. The mean of the multiple similarity scores is calculated, and the mean similarity score is set as the similarity comparison result. The similarity deviation between the similarity comparison result and a preset similarity threshold is calculated, and the infrared quality detection result is determined based on the similarity deviation, wherein the similarity deviation is inversely proportional to the infrared quality detection result.

[0050] Specifically, the weld feature set includes weld bead features, incomplete penetration features, porosity features, and undercut features. A pixel feature matrix is ​​constructed based on the weld feature set. The constructed pixel feature matrix is ​​then compared with multiple unqualified template matrices in a pre-defined template matrix library using a similarity traversal. These multiple unqualified template matrices represent different types of welding defects. The similarity score of each compared unqualified template matrix is ​​output, representing the degree of similarity between the input matrix and the template matrix.

[0051] Then, the average of the obtained similarities is calculated to obtain the mean similarity, which represents the average similarity between the input matrix and the unqualified template matrix. The similarity deviation is calculated between the mean similarity and a preset similarity threshold. The preset similarity threshold is a pre-defined value used to determine whether the similarity is high enough to identify the presence of welding defects. The similarity deviation can be calculated as either the absolute or relative value of the mean similarity to determine the difference from the preset similarity threshold. The infrared quality detection result is determined based on the similarity deviation. If the similarity deviation is small, it indicates that the input matrix is ​​similar to the unqualified template matrix, possibly indicating a welding defect. If the similarity deviation is large, it indicates that the input matrix is ​​dissimilar to the unqualified template matrix, indicating good welding quality. The similarity deviation is inversely proportional to the infrared quality detection result, meaning that the smaller the similarity deviation, the more serious the detected welding quality problem.

[0052] By comparing and evaluating welding quality, welding defects can be quickly and accurately identified by comparing the actual acquired image features with known defect templates, thereby improving detection efficiency and reducing misjudgments.

[0053] Furthermore, this application also includes: Multiple defective weld feature images are acquired, and multiple initial defective template matrices are constructed based on these images. Based on the cosine similarity principle, similarity analysis is performed on any two initial defective template matrices sequentially to obtain the minimum similarity and mean similarity of each matrix. When the minimum similarity is less than a first similarity threshold and the mean similarity is less than a second similarity threshold, the corresponding initial defective template matrix is ​​added to the pixel feature matrix, where the first similarity threshold is less than the second similarity threshold. This process continues until a preset matrix quantity constraint is met, thus completing the construction of the pixel feature matrix.

[0054] Specifically, images of multiple defective welds are collected to demonstrate the characteristics of various welding defects, such as weld beads, incomplete penetration, porosity, and undercut. Based on the collected images of defective weld features, multiple initial defect template matrices are constructed.

[0055] Then, cosine similarity, a method for measuring the difference in direction between two vectors, is used to calculate the similarity of image feature vectors. The cosine similarity principle is used to perform similarity analysis on any two initial unqualified template matrices. For each initial unqualified template matrix, the minimum similarity to other matrices and the mean similarity of all similarities are calculated.

[0056] Next, the minimum and mean similarity of each initial unqualified template matrix are compared with a preset first and second similarity thresholds. When the minimum similarity is less than the first similarity threshold and the mean similarity is less than the second similarity threshold, the corresponding initial unqualified template matrix is ​​added to the pixel feature matrix. This ensures that only new templates that are sufficiently different from existing templates are added, thereby maintaining the diversity and discriminative power of the template matrix. The first similarity threshold being less than the second similarity threshold indicates a stricter requirement for minimum similarity.

[0057] Next, the iteration is repeated until the number of unqualified template matrices in the pixel feature matrix meets the preset constraints, thereby controlling the size of the matrix to ensure computational efficiency while ensuring sufficient representativeness.

[0058] By constructing a pixel feature matrix containing various defective welding features, subsequent similarity comparisons can be used to identify and evaluate welding quality, thereby improving the accuracy and efficiency of welding defect identification.

[0059] Furthermore, this application also includes: A sample image detection dataset, a sample point cloud detection dataset, and a sample infrared detection dataset are obtained through online retrieval. Based on these datasets, image detection accuracy, point cloud detection accuracy, and infrared detection accuracy are calculated respectively. Using the coefficient of variation method, a confidence weight is set according to the image detection accuracy, point cloud detection accuracy, and infrared detection accuracy to construct a detection result fusion strategy. Based on this fusion strategy, the image quality detection result, point cloud quality detection result, and infrared quality detection result are weighted and calculated to obtain the welding quality detection result.

[0060] Specifically, sample image detection datasets, sample point cloud detection datasets, and sample infrared detection datasets are obtained through online retrieval for training and validation. These datasets contain a large amount of labeled data and are used to evaluate the accuracy of the detection methods.

[0061] Then, the image detection accuracy, point cloud detection accuracy, and infrared detection accuracy were calculated based on the sample image detection dataset, sample point cloud detection dataset, and sample infrared detection dataset, respectively, reflecting the performance of each detection method in identifying welding defects.

[0062] Next, the coefficient of variation (COP) method, a statistical approach used to measure the dispersion of data, is employed. Using COP, confidence weights are assigned based on the stability of image detection accuracy, point cloud detection accuracy, and infrared detection accuracy. A smaller COP indicates more stable accuracy, and vice versa. Higher accuracy results in higher weights, signifying more reliable detection methods, while lower accuracy indicates lower reliability. Based on these calculated confidence weights, detection results from different sources are combined to construct a detection result fusion strategy.

[0063] Next, based on the detection result fusion strategy, the image quality detection result, point cloud quality detection result, and infrared quality detection result are weighted according to the corresponding accuracy and confidence weight to obtain the welding quality detection result. By integrating information from different detection technologies, a comprehensive assessment of welding quality is provided.

[0064] By leveraging the advantages of different testing technologies and considering their respective reliability and stability, the overall accuracy and effectiveness of welding quality testing can be improved, enabling more accurate identification and resolution of welding problems.

[0065] Furthermore, this application also includes: The actual detection data within a preset time period is statistically analyzed, and the actual detection accuracy rate is calculated based on the actual detection data; the detection result fusion strategy is updated according to the actual detection accuracy rate.

[0066] Specifically, within a preset time period, actual detection data is collected and statistically analyzed, including image detection results, point cloud detection results, and infrared detection results, as well as the corresponding actual welding quality. Based on the collected actual detection data, the actual detection accuracy of each detection method is calculated by comparing the matching degree between the detection results and the actual welding quality, reflecting the performance of each detection method in practical applications.

[0067] Then, the detection result fusion strategy is updated based on the calculated actual detection accuracy. If the actual accuracy of a certain detection method is higher than expected, its weight in the fusion strategy is increased; conversely, if the actual accuracy is lower than expected, its weight is decreased. The update strategy may include adjusting the confidence weights, changing the weighting calculation formula, or introducing new detection data sources.

[0068] By ensuring that the testing system always reflects the latest testing performance and can adapt to possible changes, such as changes in the testing environment, welding process, or materials, the overall accuracy and robustness of the testing system are improved, making it more suitable for real-world application scenarios.

[0069] In summary, the BIM-based integrated prefabricated steel structure assembly quality inspection method provided in this application has the following technical advantages: Data is collected from the target welding area at the connection point using a multi-source device array to obtain a multi-source dataset, which includes a regional image set, a point cloud dataset, and an infrared thermal imaging image. The regional image set is fused according to a preset image fusion strategy to obtain a bidirectional regional fusion image. Features are extracted from the bidirectional regional fusion image using a feature recognition channel to determine the weld feature set. Welding quality is assessed based on the weld feature set to obtain image quality detection results. A contour extraction model is used to extract contours from the bidirectional regional fusion image to construct a three-dimensional regional contour. Point cloud fitting is performed on the point cloud dataset based on the three-dimensional regional contour to generate a three-dimensional regional model. Mesh defect identification is performed on the three-dimensional regional model to generate... Point cloud quality inspection results; the infrared thermal imaging image is preprocessed according to a preset image processing strategy to obtain a grayscale enhanced image, the grayscale enhanced image is identified, and the weld feature set is determined. The preset image processing strategy includes grayscale processing, histogram equalization enhancement, and binarization processing. A pixel feature matrix is ​​constructed based on the weld feature set, and the pixel feature matrix is ​​input into a preset template matrix library for similarity traversal comparison. The infrared quality inspection results are determined based on the similarity comparison results. The image quality inspection results, the point cloud quality inspection results, and the infrared quality inspection results are fused to output the welding quality inspection results of the target welding area, thereby achieving the technical goal of multi-source defect detection and improving the accuracy and efficiency of weld quality assessment at connection points.

[0070] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0071] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of this application and its equivalents, this application also intends to include such modifications and variations.

Claims

1. A BIM-based integrated prefabricated steel structure assembly quality inspection method, characterized in that, The method includes: Based on a multi-source device array, data is collected from the target welding area at the connection point to obtain a multi-source dataset, which includes a regional image set, a point cloud dataset, and an infrared thermal imaging image. The region image set is fused according to a preset image fusion strategy to obtain a bidirectional region fused image. Features are extracted from the bidirectional region fused image through a feature recognition channel to determine the weld feature set. Welding quality is evaluated based on the weld feature set to obtain the image quality detection result. The contour extraction model is used to extract the contour of the bidirectional region fusion image to construct a three-dimensional region contour. The point cloud dataset is fitted based on the three-dimensional region contour to generate a three-dimensional region model. Mesh defect identification is performed on the three-dimensional region model to generate a point cloud quality detection result. The infrared thermal imaging image is preprocessed according to a preset image processing strategy to obtain a grayscale enhanced image. The grayscale enhanced image is then identified to determine the weld feature set. The preset image processing strategy includes grayscale processing, histogram equalization enhancement, and binarization processing. A pixel feature matrix is ​​constructed based on the weld feature set. The pixel feature matrix is ​​input into a preset template matrix library for similarity traversal and comparison. The infrared quality detection result is determined based on the similarity comparison result. By fusing the image quality detection results, the point cloud quality detection results, and the infrared quality detection results, the welding quality detection results of the target welding area are output.

2. The method according to claim 1, characterized in that, The region image set is fused according to a preset image fusion strategy to obtain a bidirectional region fused image, including: The region image set is divided based on the forward angle and the lateral angle to obtain a forward region image set and a lateral region image set; Obtain a preset image fusion strategy, wherein the preset image fusion strategy includes a forward fusion strategy and a lateral fusion strategy; The forward fusion strategy is a mean fusion strategy, and the lateral fusion strategy is a weighted calculation strategy for overlapping regions, wherein the weight of the overlapping regions is inversely proportional to the image interference coefficient of the acquisition angle. The forward fusion strategy and the lateral fusion strategy are used to fuse the forward region image set and the lateral region image set respectively to obtain a forward region fused image and a lateral region fused image, which are set as the bidirectional region fused image.

3. The method according to claim 2, characterized in that, Constructing a contour extraction model includes: Using the attribute characteristics of the target welding area as constraints, a sample forward image set, a sample forward contour set, a sample lateral image set, and a sample lateral contour set are obtained through network retrieval. Using the sample forward image set and the sample forward contour set as forward training data, the convolutional neural network is trained under supervision and feedback verification to obtain a forward contour extraction unit that meets the expected convergence constraints. Using the sample lateral image set and the sample lateral contour set as lateral training data, the convolutional neural network is trained under supervision and validated with feedback to obtain a lateral contour extraction unit that meets the expected convergence constraints. The contour extraction model is constructed based on the forward contour extraction unit and the lateral contour extraction unit.

4. The method according to claim 1, characterized in that, Based on the three-dimensional region contour, point cloud fitting is performed on the point cloud dataset to generate a three-dimensional model of the region, including: Using the three-dimensional region contour as a constraint, the point cloud dataset is randomly fitted to obtain a first point cloud fitting result, and the first fitting convergence of the first point cloud fitting result is calculated, wherein the first fitting convergence is the ratio of the number of point cloud data falling within the three-dimensional region contour to the total number of point cloud data in the point cloud dataset. Again, using the three-dimensional region contour as a constraint, the point cloud dataset is randomly fitted to obtain a second point cloud fitting result and a second fitting convergence. If the first fitting convergence is greater than the second fitting convergence, then the first point cloud fitting result is taken as the current optimal point cloud fitting result; otherwise, if the first fitting convergence is less than or equal to the second fitting convergence, then the second point cloud fitting result is taken as the current optimal point cloud fitting result. Perform iterative point cloud fitting until the preset number of fitting iterations is met, and output the current optimal point cloud fitting result to construct the three-dimensional model of the region.

5. The method according to claim 1, characterized in that, Mesh defect identification is performed on the 3D model of the region to generate point cloud quality inspection results, including: Obtain preset mesh defect indices, wherein the preset mesh defect indices include crack indices and porosity indices; Based on the preset mesh defect index, mesh defect identification is performed on the three-dimensional model of the region to determine the crack feature set and the porosity feature set. Based on the crack feature set and the porosity feature set, the regional welding quality is evaluated, and the point cloud quality detection result is obtained.

6. The method according to claim 1, characterized in that, The pixel feature matrix is ​​input into a preset template matrix library for similarity comparison. The infrared quality detection result is determined based on the similarity comparison result, including: The weld feature set includes at least weld bead features, incomplete penetration features, porosity features, and undercut features; A pixel feature matrix is ​​constructed based on the weld feature set, and the pixel feature matrix is ​​compared with multiple unqualified template matrices in the preset template matrix library to output multiple similarity scores. The average of the multiple similarities is calculated, and the average similarity is set as the similarity comparison result; The similarity deviation between the similarity comparison result and the preset similarity threshold is calculated, and the infrared quality detection result is determined based on the similarity deviation, wherein the similarity deviation is inversely proportional to the infrared quality detection result.

7. The method according to claim 6, characterized in that, Constructing the pixel feature matrix includes: Multiple defective weld feature images are acquired, and multiple initial defective template matrices are constructed based on the multiple defective weld feature images; Based on the cosine similarity principle, similarity analysis is performed on any two initial unqualified template matrices from the plurality of initial unqualified template matrices in turn to obtain the minimum similarity and mean similarity of each initial unqualified template matrix. When the minimum similarity is less than the first similarity threshold and the mean similarity is less than the second similarity threshold, the corresponding initial unqualified template matrix is ​​added to the pixel feature matrix, wherein the first similarity threshold is less than the second similarity threshold; The construction of the pixel feature matrix is ​​completed until the preset matrix quantity constraint is met.

8. The method according to claim 1, characterized in that, The fusion of the image quality detection result, the point cloud quality detection result, and the infrared quality detection result includes: The sample image detection dataset, sample point cloud detection dataset, and sample infrared detection dataset are retrieved via network search. The image detection accuracy, point cloud detection accuracy, and infrared detection accuracy are calculated based on the sample image detection dataset, the sample point cloud detection dataset, and the sample infrared detection dataset, respectively. Using the coefficient of variation method, a confidence weight is set according to the image detection accuracy, the point cloud detection accuracy, and the infrared detection accuracy to construct a detection result fusion strategy; The image quality detection result, the point cloud quality detection result, and the infrared quality detection result are weighted and calculated based on the detection result fusion strategy to obtain the welding quality detection result.

9. The method according to claim 8, characterized in that, The method further includes: Collect actual detection data within a preset time period, and calculate the actual detection accuracy rate based on the actual detection data; The detection result fusion strategy is updated based on the actual detection accuracy.

Images