Method for repairing defective part of building component

Abstract

The invention discloses a method for repairing a defective part of a building component. The method comprises the steps of collecting images of the defective building component from multiple angles; establishing a three-dimensional model diagram of the defective building component and extracting a contour line of the three-dimensional model diagram; identifying the facing type of the defective building component through multiple models; searching a component model closest to the modeling style and the facing type of the defective building component in a component model database by utilizing the trained deep learning model according to the facing type and the contour line, and taking the component model as a repair sample; according to the modeling style of the repair sample, the defect part in the three-dimensional model diagram of the defect building component is subjected to vacancy repair to form a three-dimensional diagram of the defect part; defective components are printed and manufactured through the 3D printing technology; mounting the defective part at the defective part of the building component; and the outer surface of the defective part is subjected to repairing construction of a facing material according to the facing type of the repairing sample. According to the method, the modeling style of the building component and the repairing quality of the facing type can be improved, and the overall modeling and facing decoration effect is guaranteed.

Description

Technical Field

[0001] The invention relates to the field of building maintenance, and in particular to a method for repairing a defective part of a building component. Background Art

[0002] In building structures, as the service life increases, the various building components in the building structure may have structural and decorative defects. At present, the detection of the structural modeling and decorative type of the defective parts of building components is carried out manually, and its detection efficiency is low. The judgment of the style type of the structural modeling and the material of the decorative type of the defective parts relies on construction experience, and misjudgment often occurs. After the defective parts of the building components are repaired according to the results of the misjudgment, the matching effect of the structural modeling style and decorative type of the repaired parts with the original building components is poor, and there are errors, which affects the overall modeling and decorative effect of the existing building. / / The repair of the defective parts of the existing building is usually filled, which leads to quality defects such as cracks between the filled area and the original building, affecting the decorative effect of the existing building surface. Summary of the invention

[0003] The purpose of the present invention is to provide a method for repairing defective parts of building components, so as to solve the problem that manual detection of structural modeling and finishing type of defective parts of building components is inefficient and has large errors, which affects the overall structural modeling and finishing decoration of the defective parts of building components after repair.

[0004] In order to solve the above technical problems, the present invention provides a method for repairing a defective part of a building component, comprising: Collect images of defective building components containing defective parts from multiple angles; Establishing a three-dimensional model diagram of the defective building components based on images of the defective building components collected from multiple angles; Extract the contour lines of the three-dimensional model of the defective building components; Using the trained multi-model to identify the finishing type of the defective building components in the collected images of the defective building components; According to the finish type and the contour line of the three-dimensional model of the defective building component, the trained deep learning model is used to search the component model database for the closest modeling style and finish type to the defective building component as a repair sample; According to the modeling style of the repair sample, the defective parts in the three-dimensional model of the defective building component are repaired to form an independent three-dimensional map of the defective parts; The defective part is produced by printing the three-dimensional image of the defective part through 3D printing technology; Installing the defective component at the defective part of the building component; Repair the outer surface of the damaged part with finishing materials according to the finishing type of the repair sample.

[0005] Furthermore, for the method for repairing defective parts of building components provided by the present invention, images of defective building components containing defective parts are collected from multiple angles by a high-precision camera carried by a drone.

[0006] Furthermore, for the method for repairing defective parts of building components provided by the present invention, the method for establishing a three-dimensional model diagram of a defective building component based on images of the defective building component collected from multiple angles includes: analyzing and processing the collected images of the defective building component through the SFM algorithm to establish a three-dimensional model diagram of the defective building component.

[0007] Furthermore, for the method for repairing defective parts of building components provided by the present invention, the method for identifying the finish type of a defective building component from the collected images of the defective building component through a trained multi-model includes: Judging the finish type of the defective building component separately from the collected images of the defective building component through the trained multi-model; When the judgment results of each model on the finish type are the same, determining this judgment result as the finish type of the building component; When the judgment results of each model are different, according to the probability fusion formula (1), taking the one with the higher probability in the judgment results as the finish type of the building component; (C i )= (1) In formula (1), C i is the finish type, (C i ) is the probability that the image of the building component containing the defective part is of finish type C i , is the probability output of the kth model for the finish category , is the weight of the kth model.

[0008] Furthermore, for the method for repairing defective parts of building components provided by the present invention, the trained multi-model includes a main model and an auxiliary model, the main model is YOLOv8, and the auxiliary model is EfficientNet-B4.

[0009] Furthermore, for the method for repairing defective parts of building components provided by the present invention, the method for extracting the contour line of the three-dimensional model diagram of the defective building component includes: Converting the three-dimensional model diagram of the defective building component into a two-dimensional image, and detecting the edge contour of the two-dimensional image through an edge detection algorithm; Performing morphological operations on the detection result of the edge contour to remove noise and broken edges and connect discontinuous contour lines; Extract the contour line of the defective building component through Fourier descriptors and identify the styling style of its shape.

[0010] Compared with the prior art, the beneficial effects of the building component defect repair method provided by the present invention are as follows: Identify the finish type of the defective building component through the images of the defective building component collected by multiple models, thereby improving the accuracy of detecting the finish type of the defective building component and reducing the judgment error of the finish type.

[0011] Using the contour line and finish type of the three-dimensional model diagram of the defective building component as inputs, and using the trained deep learning model to find the component model in the component model database that is closest to the styling style and finish type of the contour line shape of the defective building component as a repair example, thereby determining the styling style of the defective building component, improving the detection efficiency and detection accuracy of the styling style of the defective building component, and reducing the error of manual detection and judgment.

[0012] Complete the repair of the defective part in the three-dimensional model diagram of the defective building component through the styling style of the repair example to form a three-dimensional diagram of the defective part that can be removed from the three-dimensional model diagram of the defective building component. Print the three-dimensional diagram of the defective part through 3D printing technology to produce the defective component, thereby improving the production quality and production accuracy of the defective part of the building component. Install it on the building component to make the two perfectly match, improving the repair quality and repair accuracy of the defective part, and ensuring the overall styling style of the repaired building component.

[0013] Carry out the repair construction of the finish material on the outer surface of the defective component through the finish type of the repair example, so that the repaired finish material is consistent with or closest to the original finish material of the defective building component to the greatest extent, thereby ensuring the overall finish decoration effect of the repaired building component. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 is a flowchart of the building component defect repair method. DETAILED DESCRIPTION OF THE INVENTION

[0015] The present invention will be described in detail below with reference to the accompanying drawings: According to the following description, the advantages and features of the present invention will be clearer. It should be noted that the accompanying drawings are all in a very simplified form and use non-precise scales, only for the purpose of facilitating and clearly assisting in explaining the purpose of the embodiments of the present invention.

[0016] Please refer to Figure 1 , an embodiment of the present invention provides a building component defect repair method, which may include: Step S1, collect images of the defective building component with the defective part from multiple angles. To improve the collection accuracy and convenience, a high-precision camera can be carried by a drone to capture photos of the defective building component with the defective part from multiple angles for image collection.

[0017] Step S2, establish a 3D model diagram of the defective building component based on the images of the defective building component collected from multiple angles. Specifically, it can include: analyzing and processing the collected images of the defective building component through the SFM algorithm to establish a 3D model diagram of the defective building component. Among them, SFM (Structure from Motion) is a computer vision technology that restores the 3D geometric structure of the scene and the camera motion trajectory based on multi-view 2D image sequences. Its technical route is to analyze the feature matching relationship and geometric constraints between images, and jointly optimize the pose (position and attitude) of the camera and the 3D point coordinates of the scene to achieve 2D to 3D modeling.

[0018] Step S3, extract the contour line of the 3D model diagram of the defective building component. Specifically, it can include: converting the 3D model diagram of the defective building component into a 2D image, and detecting the edge contour of the 2D image through an edge detection algorithm, such as the Canny edge detection algorithm; to improve the accuracy and integrity of the contour line, morphological operations (dilation, erosion) can be performed on the detection result of the edge contour to remove noise and broken edges and connect discontinuous contour lines; then extract the contour line of the defective building component through Fourier Descriptors. Thus, the styling style of the defective building component can be identified through the contour line.

[0019] Step S4, identify the finish type of the defective building component from the collected images of the defective building component through the trained multi-model. The finish types include but are not limited to pebble dash, terrazzo, paint, gold leaf finish, etc. Among them, the trained multi-model can use YOLOv8 as the main model and EfficientNet-B4 as the auxiliary model. When training the two models, use the labeled samples of building finishes such as pebble dash, terrazzo, paint, gold leaf finish, etc. to train, and focus on optimizing the detection ability of small targets (such as the details of gold leaf finish). Thus, improve the accuracy of the judgment results of each model. The auxiliary model can output the global class probability (such as gold leaf finish 0.95, paint 0.03), which is used to supplement the global finish type information after the main model identifies the finish type of the defective building component. When the judgment results of each model on the finish type are the same, determine this judgment result as the finish type of the building component; when the judgment results of each model are different, according to the probability fusion formula (1), take the one with the higher probability in the judgment results as the finish type of the building component.

[0020] (C i ) = (1); In formula (1), C i is the finish type, (C i ) is the probability of judging that the image of the building component with a defect part is of the C i finish type, is the probability output of the k-th model for the finish category , is the weight of the k-th model. The weights are assigned according to the accuracy of each model on the validation set, where the weight of YOLOv8 can be 0.6 and the weight of EfficientNet-B4 can be 0.4.

[0021] If all models output a probability ( i ) > 0.8) for a certain finish type C , then the judgment result of the finish material of the output image is C i . If there are differences (such as 0.6 for gold leaf finish vs. 0.5 for paint finish), the maximum value is taken according to the weighted probability. If the weighted probabilities of multiple finishes are relatively close (the probability difference is less than 0.01), manual intervention is required to judge the finish type C i .

[0022] To improve the recognition accuracy, before the multi-model recognizes the finish type, the collected images can be enhanced. For example: using contrast-limited adaptive histogram equalization (CLAHE) to enhance the contrast and clarity of the images and highlight the texture and color features of the finish.

[0023] The order of steps S2, S3 and step S4 can be interchanged.

[0024] Step S5, according to the finish type and the contour line of the three-dimensional model diagram of the defective building component, use the trained deep learning model to search in the component model database for the component model that is closest to the styling style and finish type of the defective building component as the repair example. In the component model database, a large number of component models with different finish types and styling styles and their corresponding feature vectors are stored, including the finish material type, length, width, perimeter, symmetry, regularity, curvature change rate, and the size of the area enclosed by the contour line. Calculate the similarity between the feature vector of the component to be repaired and the feature vectors of each component model in the database through cosine similarity, and find the component model with the highest similarity as the component model of the similar type as the repair example. The trained deep learning model can be BERT.

[0025] Step S6, fill and repair the defective parts in the three-dimensional model of the defective building component according to the modeling style of the repair sample to form a three-dimensional map of the defective part. According to the actual use environment and requirements of the defective building component, a material with suitable strength, durability and appearance characteristics can be selected. For example, for components that bear a large load, high-strength plastic or metal materials can be selected; for components with high requirements for appearance, resin materials with good surface quality can be selected.

[0026] Step S7, the defective part is produced by printing the three-dimensional image of the defective part through 3D printing technology. When designing the 3D printing model, its connection surface is designed to be slightly larger than the three-dimensional image of the defective part, generally exceeding the edge of the defective part by 10-20mm. This ensures that it can be better connected and fixed with the original component during installation.

[0027] Step S8, installing the defective part at the defective part of the building component. The specific installation method may include: selecting a suitable connection method according to whether the angle of the defective part is close to a vertical plane or a horizontal plane. If the defective part is close to a vertical plane, a bolt + glue connection method can be used. First, drill holes on the original component and the defective part made by 3D printing respectively, and then use bolts to preliminarily fix the two, and then apply high-strength glue on the connection part to enhance the stability and sealing of the connection. If the defective part is located on a horizontal plane, a hidden step mouth + bolt connection method can be used. A hidden step mouth is machined on the original component, and the corresponding part of the defective part made by 3D printing is embedded in the step mouth, and then fixed with bolts. This connection method can improve the strength and stability of the connection, while making the connection part more concealed and not affecting the appearance of the component.

[0028] Step S9, repairing the outer surface of the defective component with a finishing material according to the finishing type of the repair sample.

[0029] The method for repairing defective parts of building components provided in an embodiment of the present invention uses a multi-model to identify the finishing type of the defective building components from the collected images of the defective building components, thereby improving the accuracy of detecting the finishing type of the defective building components and reducing the judgment error of the finishing type.

[0030] The method for repairing defective parts of building components provided in an embodiment of the present invention uses the contour line and finish type of the three-dimensional model of the defective building component as input, and uses the trained deep learning model to search the component model database for the modeling style and finish type that are closest to the contour line shape of the defective building component as a repair sample, thereby determining the modeling style of the defective building component, improving the detection efficiency and detection accuracy of the modeling style of the defective building component, and reducing the errors of manual detection and judgment.

[0031] The building component defect repair method provided by the embodiments of the present invention complements and repairs the defect parts in the three-dimensional model diagram of the defective building component through the styling style of the repair sample to form a three-dimensional diagram of the defect parts that can be removed from the three-dimensional model diagram of the defective building component. The defective parts are manufactured by 3D printing the three-dimensional diagram of the defect parts, thereby improving the manufacturing quality and precision of the defective parts of the building component. Installing it on the building component makes the two perfectly match, improving the repair quality and precision of the defective parts, and ensuring the overall styling of the repaired building component.

[0032] The building component defect repair method provided by the embodiments of the present invention performs the repair construction of the finishing material on the outer surface of the defective part through the finishing type of the repair sample, so that the repaired finishing material is consistent with or maximally close to the original finishing material of the defective building component, thus ensuring the overall finishing decoration effect of the repaired building component.

[0033] The building component defect repair method provided by the embodiments of the present invention can realize the full-process intelligence and precision of building component defect repair, effectively improve the repair efficiency and quality, and ensure that the repaired component highly fits the original component in terms of structural strength and appearance style.

[0034] The present invention is not limited to the above specific embodiments. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of them. Based on the described embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art belong to the scope of protection of the present invention. Those skilled in the art can make other levels of modifications and changes to the present invention. Thus, if these modifications and changes of the present invention are within the scope of the claims of the present invention, the present invention also intends to include these modifications and changes.

Claims

1. A method for repairing a defective part of a building component, characterized in that: include: Collect images of defective building components containing defective parts from multiple angles; Establishing a three-dimensional model diagram of the defective building components based on images of the defective building components collected from multiple angles; Extract the contour lines of the three-dimensional model of the defective building components; Using the trained multi-model to identify the finishing type of the defective building components in the collected images of the defective building components; According to the finish type and the contour line of the three-dimensional model of the defective building component, the trained deep learning model is used to search the component model database for the closest modeling style and finish type to the defective building component as a repair sample; According to the modeling style of the repair sample, the defective parts in the three-dimensional model of the defective building component are repaired to form an independent three-dimensional map of the defective parts; The defective part is produced by printing the three-dimensional image of the defective part through 3D printing technology; Installing the defective component at the defective part of the building component; Repair the outer surface of the damaged part with finishing materials according to the finishing type of the repair sample.

2. The method for repairing a defective part of a building component according to claim 1, characterized in that: Use drones equipped with high-precision cameras to collect images of defective building components containing defective parts from multiple angles.

3. The method for repairing a defective part of a building component according to claim 1, characterized in that: The method for establishing a three-dimensional model diagram of a defective building component based on images of defective building components collected from multiple angles includes: analyzing and processing the collected images of defective building components through an SFM algorithm to establish a three-dimensional model diagram of the defective building component.

4. The method for repairing a defective part of a building component according to claim 1, characterized in that: The method for identifying the finishing type of the defective building component by using the trained multi-model for the collected image of the defective building component includes: The images of defective building components collected are used to determine the finishing types of the defective building components through the trained multi-models; When the judgment results of the models on the finishing type are the same, the judgment result is determined as the finishing type of the building component; When the judgment results of each model are different, according to the probability fusion formula (1), the one with the highest probability in the judgment result is taken as the finishing type of the building component; (C i )= （1） In formula (1), C i Is the type of finish, (C i ) is used to judge the image of the building component with defective parts as C i Probability of facing type, is the kth model for the facing category The probability output is is the weight of the k-th model.

5. The method for repairing a defective part of a building component according to claim 4, characterized in that: The trained multi-model includes a main model and an auxiliary model, the main model is YOLOv8, and the auxiliary model is EfficientNet-B4.

6. The method for repairing a defective part of a building component according to claim 1, characterized in that: The method for extracting the contour line of the three-dimensional model of the defective building component includes: Convert the three-dimensional model of the defective building component into a two-dimensional image, and detect the edge contour of the two-dimensional image through an edge detection algorithm; Perform morphological operations on the edge contour detection results to remove noise and broken edges and connect discontinuous contour lines; The contours of the missing building components are extracted through Fourier descriptors and the modeling style of their shapes is identified.

Images