Reinforcing steel bar concealed engineering intelligent positioning method based on ground penetrating radar and machine learning
Through ground penetrating radar and machine learning technology, the problems of low efficiency and poor objectivity of steel bar detection are solved, intelligent positioning and health monitoring of steel bar hidden projects are realized, construction quality and information management are improved, and mechanical performance of subway facilities is ensured. and durability.
Patent Information
- Application Number
- CN202411012636.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-26
- Publication Date
- 2024-11-15
AI Technical Summary
In the construction of steel bar engineering, it is difficult for the existing technology to efficiently and reliably detect the characteristic data of steel bars, resulting in low detection efficiency and poor objectivity, and lack of sharing of overall health monitoring information for in-service building components, with cutting corners, layout deviations and steel bars Problems such as pedaling deformation affect the mechanical properties and durability of subway facilities.
The ground penetrating radar combined with machine learning technology is used to extract the spatial data of steel bars through denoising processing and deep learning algorithms, and the MaskRCNN model is used to segment the steel bar target features, and the curve is fitted through the least squares method, and three-dimensional reconstruction is combined with BIM technology to realize the hidden steel bars. Intelligent positioning and health monitoring of the project.
Improve the efficiency and reliability of steel bar inspection, reduce the dependence on manual experience, improve the construction quality and information management level, ensure the unity and safety of construction quality, and realize the overall health monitoring information of in-service building components shared.
Smart Images

Figure CN118962669A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to an intelligent positioning method for concealed steel bar engineering based on ground penetrating radar and machine learning, belonging to the technical field of engineering detection. Background Art
[0002] In the construction of steel bar projects, there may be situations such as cutting corners, layout deviations, and deformation of steel bars due to trampling, which results in the thickness of the steel bar protective layer and the steel bar spacing not meeting the requirements of relevant specifications, greatly reducing the mechanical properties and durability of the steel-concrete structure, and seriously affecting the normal service of subway facilities.
[0003] At the same time, subway projects are large in construction volume and wide in area, and it is impossible to discover such steel bar construction defects in time for quality management. Moreover, the quality acceptance of hidden steel bar projects is usually carried out before the pouring is completed. Once the pouring is completed, the internal structure cannot be restored, which poses a major safety hazard to the overall project quality.
[0004] At present, in the task of detecting in-service steel bars in steel-concrete structures, professionals engaged in ground-penetrating radar detection mainly find detection targets from complex echo signals based on their accumulated professional knowledge and previous work experience. In the task of detecting in-service steel bars in large-scale structures such as subways, the collected ground-penetrating radar data is often huge. Manually extracting the characteristic data of steel bars from a large amount of data is time-consuming and labor-intensive, and the detection efficiency is low. The objectivity of the data is poor, and it is difficult to meet the needs of modern quality inspection. Traditional data relies on manual archiving, which is prone to misdetection or missed detection, and the existing technology lacks the sharing of overall health monitoring information of in-service building components.
[0005] Therefore, there is currently a lack of inventions that address the needs of improving the efficiency, reliability and objectivity of ground penetrating radar steel bar target detection tasks. Summary of the invention
[0006] The purpose of the present invention is to provide an intelligent positioning method for hidden steel bar projects based on ground penetrating radar and machine learning. The internal data of the steel-concrete structure is obtained by means of ground penetrating radar non-destructive testing technology. The original ground penetrating radar data is denoised and then the steel bar spatial data is extracted and classified and stored in an informationized manner through deep learning technology. Finally, the extracted steel bar spatial data is reversely reconstructed in three dimensions to realize the sharing of the overall health monitoring information of the in-service building components.
[0007] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for intelligent positioning of hidden steel bar engineering based on ground penetrating radar and machine learning, comprising the following steps: (1) Real-time data collection: After the main facilities of the subway station are completed, the SIR-3000 ground penetrating radar is used to collect steel bar radar wave data; (2) Data preprocessing: A denoising algorithm is used to remove the interference noise of the direct wave noise and the ground reflection wave in the collected steel bar radar wave data. The corresponding images are extracted into the training set and test set files through the program. The extracted corresponding image contents include: the original image of the ground penetrating radar data, labels, masks, and types; at the same time, a file path list is generated to provide path guidance for the subsequent extraction of feature data; (3) Steel bar positioning detection: OpenCV2 is used to read and write ground penetrating radar images, and the input image information is sent to the MaskRCNN model to segment and extract the hyperbolic features of the steel bar target. The least squares method is used for curve fitting to obtain the hyperbola and extract the coordinate points of the vertices. The spatial data is output according to the relative position of the steel bar in the image. (4) Classify and store steel bar data: According to the different characteristics of steel bar protective layer thickness in different structures, determine the steel bar radar wave data corresponding to the plate, column and wall respectively, generate and save them in the corresponding data files; use ground penetrating radar images to generate time intervals, determine the transverse and longitudinal bars, generate and save them in the corresponding data files; (5) Separation of digital and model: The Dynamo parametric tool is used to perform three-dimensional geometric projection of the steel bar spatial data in the data file. The steel bar spatial data is combined with BIM technology to reversely reconstruct the steel bars in the steel-concrete structure in the subway project. The steel bar spatial data is mapped to the model to form a visual GPR to BIM model, realizing the sharing of the overall health monitoring information of the in-service building components.
[0008] In the preferred technical solution, in step (3), the hyperbolic features of the steel bar target are segmented and extracted in the MaskRCNN model as follows: an algorithm with a Mask RCNN network with RseNet101 as the backbone network as the core is adopted, and the SENet module is integrated to perform steel bar target detection.
[0009] In the preferred technical solution, the Mask RCNN network model first uses the ResNet101 network as the backbone network to input the image, then extracts the features of the image, and constructs the feature pyramid structure using the feature measurement of the fixed ratio of length and width compression; then uses the FPN network to extract multi-scale target features, using deep features for large targets and shallow features for small targets; in the RNP network of the Mask RCNN network model, a sliding window is used to scan the image, and the scanned areas are classified to extract candidate areas with high scores, and then the shared features of the candidate boxes of the RoI Align layer are input for extraction, and finally the tasks of target detection, instance segmentation and mask prediction are completed.
[0010] In a further technical solution, the SENet module includes three processes: squeezing, excitation, and weighting. The original data X is convolved to obtain U, and the data U is input into the SENet module, where H and W represent the length and width of a channel, respectively.
[0011] In a further technical solution, the SENet module includes a Squeeze module 、Excitation module With the Scale module First, after the data U is input into the SENet module, it passes through the compression module The module converts each feature map into a standard quantity through a global average pooling operation. The converted data is The matrix , improve the global receptive field of the data, and its calculation formula is shown in formula (1): (1) Where: is the cth channel in the feature map, is the characteristic data of the i-th row and j-th column in the channel; After completing the compression task, in the incentive module First, the matrix is transformed into A dimensionality reduction operation is performed to convert the original dimension from C to C / r. After that, it is activated by a ReLU activation function and then restored to its original dimension by a fully connected layer. Finally, the weights of all connected layers are calculated by the Sigmoid function, and the correlation between channels is learned by using the weights. Module output The calculation formula is shown in formula (2): (2) Where: Sigmoid and ReLU represent two activation functions, W1 and W2 represent two convolutional layers, The Scale module in the SENet module performs a weighted operation, which multiplies the stimulated weights by their corresponding features, redistributes the feature weights, and updates the feature map, thereby enhancing important features and suppressing unimportant features.
[0012] In a further technical solution, the feature pyramid structure in the ResNet101 network is integrated with an attention mechanism. The ResNet101 network uses a fixed length-width compression ratio, and compresses the length and width twice, three times, four times, and five times respectively to obtain feature maps C2, C3, C4, and C5, and the attention mechanism is added before the C1-C5 convolutional layers.
[0013] In the preferred technical solution, the output of spatial data according to the relative position of the steel bars in the picture is: substituting the coordinate points of the extracted vertices into the calculation formula of the steel bar spacing (L1) and the steel bar protective layer thickness (H1), and the calculation formula is as follows: (3) (4) Where: H is the total detection depth, L is the detection length, H y is the width of the GPR image, L x is the length of the GPR image, x and y are the relative coordinates of the central vertex of the hyperbola in the GPR image, and the spatial data of the spacing between the steel bars and the thickness of the protective layer are obtained and stored in Excel.
[0014] In the preferred technical solution, in step (3), the least squares method is used to curve fit the steel mask segmented from Mask RCNN, and the coordinate points of the vertices are extracted. The output x and y are respectively substituted into the formula (3) and (4) Calculate the spatial data including the spacing between steel bars and the thickness of the protective layer.
[0015] In the preferred technical solution, the program in step (2) is a Python program.
[0016] In the preferred technical solution, the data files in steps (4) and (5) are Excel files.
[0017] The steel bar radar wave data collected in the present invention contains images in the form of hyperbolas. The imaging characteristics of ground penetrating radar data, whether it is the steel bar target or the interference of some non-target objects, are all presented in the form of hyperbolas, which will have some impact on the segmentation performance of MaskRCNN. An attention mechanism is introduced into the feature pyramid structure, and the characteristics of the attention mechanism are used to extract key information in the feature map, thereby improving the robustness of the network.
[0018] In the preferred technical solution, the reverse reconstruction of the steel bars in step (5) is specifically as follows: importing steel bar data from Excel into Dynamo, classifying and storing the spacing and protective layer thickness of the steel bars, extracting and outputting the spacing and protective layer thickness information of the steel bars into two different lists, set as list S and list P, using the parameters in the list to construct symmetrical points on the upper and lower surfaces of the wall, and then connecting the corresponding points inside the structure to form a straight line, and finally giving the straight steel bar entities and attributes and completing the interaction with the concrete model to form a three-dimensional model of the entire steel-concrete structure.
[0019] The above scheme is further as follows: use the "File Path" editing node in Dynamo to select the Excel file that stores the steel bar position information of the corresponding component, complete the preliminary import of the steel bar parameter information, and then read the data from the Excel table through the "Data.lmportExcel" editing node. When extracting and saving the steel bar parameters, the spacing and protective layer thickness of the steel bars are stored in the first and second columns of the corresponding Excel file respectively; at the same time, the editing node "List.GetltemAtindex" returns the data of the specified index column of the input table, and uses the "List.GetltemAtindex" editing node to read the data of the specified index column and generate a new list and output it, so as to extract and output the steel bar spacing and protective layer thickness information in two different lists, set as S and list P.
[0020] Use the above parameters to construct symmetrical points on the upper and lower surfaces of the wall, then connect the corresponding points inside the structure to form a straight line, and finally give the straight steel bar entities and attributes and interact with the concrete model to form a three-dimensional model of the entire steel-concrete structure.
[0021] The coordinate point at the lower right of the structure is output through the "Suface.PointAtParameter" editing node, and then the specific values of the point on the X, Y, and Z coordinate axes can be obtained using "Point.X", "Point.Y", and "Point.Z" respectively; then the list of steel bar spacing, protective layer thickness, etc. output by the "List.GetltemAtindex" editing node is input into "List.Map", and a "Formula" editing node is created at the other input end of "List.Map", and the values in the list representing steel bar spacing, protective layer thickness, etc. are calculated with the corresponding x, y, z to obtain a new list, which stores the coordinate value of a certain end point of the steel bar on the X, Y, and Z coordinate axes; Taking the longitudinal reinforcement as an example, assume that any end of the reinforcement It is easy to know that the longitudinal reinforcement of the wall is perpendicular to the (x, y) plane, so the x and y values of its two end points are the same, and z can be calculated based on the end points on the upper and lower bottom surfaces of the structure, that is; (5) (6) (7) In the above formula: x, y, z are the coordinates of the horizontal axis, vertical axis and vertical axis of the wall reference point respectively, P i is the protective layer thickness of a certain wall longitudinal reinforcement, S iis the spacing of the longitudinal reinforcement of a wall, and h is the height of the wall. For the transverse reinforcement in the wall, the above principle can also be used. It is only necessary to determine the plane perpendicular to the reinforcement and then deduce the calculation formula. Similarly, the endpoints of the reinforcement inside the floor slab can also be parametrically created in the same way.
[0022] In summary, the coordinate values of both ends of the steel bar can be obtained, and then the two points can be connected to generate a straight line through the "Line.ByStartPointEndPoint" editing node, and then the output result of the node "Line.ByStartPointEndPoint" is input into the "Rebar.FreeFormRebarByCurves" editing node to determine the spatial properties of the steel bar. At the same time, the size and type of the steel bar are selected by "RebarBarType" and input into the "Rebar.FreeFormRebarByCurves" editing node as the output result to determine other properties of the steel bar. Finally, run the parametric modeling results of Dynamo, create the 3D model of the steel bar into the model file of the opened built concrete structure, complete the combination of the two, and realize the creation of a complete model of the steel-concrete structure.
[0023] Due to the application of the above technical solution, the present invention has the following advantages compared with the prior art: The present invention uses a machine learning algorithm to detect and segment the hyperbolic features of steel bars in ground penetrating radar data, automatically identifies steel bar radar wave data, enhances the objectivity and reliability of the detection results, improves the efficiency of steel bar concealed engineering detection, and reduces reliance on manual experience.
[0024] Through machine learning analysis, the present invention saves the spatial data of steel bars according to the category of their embedded components, and archives the results in an informationized manner, thereby improving the overall construction quality and project productivity of steel bar concealed projects and improving the level of information management of construction sites.
[0025] The present invention replaces manual inspection of the quality of hidden steel bar projects, which can not only reduce the workload, but also avoid the inspection being affected by the limitations of the personal experience of operators and supervisors, resulting in uneven quality acceptance levels. The formulation of a unified quality evaluation standard for hidden steel bar projects can effectively ensure the construction quality.
[0026] The present invention automatically and independently stores the non-destructive testing results of steel bar components in an Excel table, and presents them in the form of a BIM digital model using Dynamo parametric modeling. It has a high degree of informatization, strong real-time performance, and outstanding sharing. BRIEF DESCRIPTION OF THE DRAWINGS
[0027] Figure 1 This is the overall flow chart of the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning; Figure 2 This is a schematic diagram of the SENet module; Figure 3 This is a schematic diagram of the Mask RCNN network structure; Figure 4 It is a schematic diagram of the residual network learning structure; Figure 5 This is a schematic diagram of the feature pyramid that integrates the attention mechanism; Figure 6 It is a schematic diagram of steel bar spatial position extraction; Figure 7 It is a schematic diagram of hyperbola fitting; Figure 8 This is a schematic diagram of the automated modeling process for wall reinforcement; Fig. 9 It is a schematic diagram of the integrated BIM model of reinforced structure and concrete structure. DETAILED DESCRIPTION
[0028] The present invention will be further described below in conjunction with the accompanying drawings and embodiments: Example 1: See Figure 1 As shown in Figure 1, an intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning is shown in Figure 1. Figure 1 As shown, the following steps are included: (1) Real-time data collection: After the main facilities of the subway station are completed, the SIR-3000 ground penetrating radar of Geophysical Survey Systems Ins. with a 900MHz antenna is used to collect steel bar radar wave data. The steel bar radar wave data contains images in the form of hyperbolas. (2) Data preprocessing: A denoising algorithm is used to remove the interference noise of the direct wave noise and the ground reflection wave in the collected steel bar radar wave data. The corresponding images are extracted into the training set and test set files through the Python program. The extracted corresponding image contents include: the original image of the ground penetrating radar data, labels, masks, and types; at the same time, a file path list is generated to provide path guidance for the subsequent extraction of feature data; (3) Steel bar positioning detection: OpenCV2 is used to read and write ground penetrating radar images, and the input image information is sent to the MaskRCNN model to segment and extract the hyperbolic features of the steel bar target. The least squares method is used for curve fitting to obtain the hyperbola and extract the coordinate points of the vertices. The spatial data is output according to the relative position of the steel bar in the image. (4) Classify and store steel bar data: According to the different characteristics of steel bar protective layer thickness in different structures, determine the steel bar radar wave data corresponding to the plate, column and wall respectively, generate and save them in the corresponding Excel file; use the ground penetrating radar image to generate the time interval, determine the transverse and longitudinal bars, generate and save them in the corresponding Excel file; (5) Separation of digital and model: The Dynamo parametric tool is used to perform three-dimensional geometric projection of the steel bar spatial data in the Excel file. The steel bar spatial data is combined with BIM technology to reversely reconstruct the steel bars in the steel-concrete structure in the subway project. The steel bar spatial data is mapped to the model to form a visual GPR to BIM model, realizing the sharing of the overall health monitoring information of the in-service building components.
[0029] In step (2), since the initial ground penetrating radar data of the subway steel-concrete structure collected often contains some clutter signals, including system noise caused by equipment reasons and other clutter caused by the inhomogeneity of the medium inside the structure and environmental factors, these interfering data are eliminated during the preprocessing process.
[0030] By writing Python programs, the corresponding images of the training set and test set are extracted into the "train_dataset" and "tset_dataset" files respectively. The main extracted information includes: original ground penetrating radar data (imgs), labels (label), masks (mask), and types (yaml); a file path list is generated based on vid_name and label to provide path guidance for subsequent feature data extraction.
[0031] The model was built using the Tensorflow2.4.0 machine learning framework. Adaptive Moment Estimation (ADAM) was selected as the optimizer. The value of Nesterov Momentum was set to 0.9, and the initial learning rate (Learning Rate, lr) was set to 0.01. The cross entropy function was used as the loss function for gradient backpropagation, and its weight decay value was set to 0.0001. The parameters in the network were updated by backpropagating the error by verifying whether the output classification result was consistent with the dataset label. The calculation results of the model's recall rate (Recall), precision rate (Precision), average precision (Average Precision, AP), mean average precision (Mean Average Precision, mAP) and F1-Score are shown in Table 1.
[0032] Table 1 Calculation results of steel bar identification model accuracy In this embodiment, the Mask RCNN algorithm in step (3) is used to segment and extract the hyperbolic features of the steel bar target. Specifically, the algorithm is used with the Mask RCNN network with RseNet101 as the backbone network as the core, and the SENet module is integrated to perform steel bar target detection.
[0033] In this embodiment, Figure 3 As shown in the figure, the Mask RCNN network model first uses the ResNet101 network as the backbone network to input the image, then extracts the image features, and uses the feature measurement of the fixed ratio of length and width compression to construct the feature pyramid structure; then the FPN network is used to extract multi-scale target features, large targets use deep features, and small targets use shallow features; in the RNP network of the Mask RCNN network model, a sliding window is used to scan the image, and the scanned area is classified. The scanned area is called an anchor, and the obtained anchor is classified to extract the candidate area with a high score, and then the shared features of the candidate box of the RoI Align layer are input for extraction, and finally the tasks of target detection, instance segmentation and mask prediction are completed.
[0034] In order to solve the degradation problem and gradient vanishing problem of traditional neural network, the residual network has many branches, such as Figure 4 As shown, the input data is directly connected to the subsequent layer, so that the subsequent layer can directly learn the residual. It is this link that directly bypasses the input information to the output that improves the problems of information loss and loss caused more or less when traditional neural networks transmit information, and protects the integrity of the information. Moreover, for the network, it only needs to learn the difference between the input and output, which simplifies the difficulty of learning. In this embodiment, the SENet module includes three processes: squeezing, excitation, and weighting. The original data X is convolved to obtain U, and the data U is input into the SENet module, where H and W represent the length and width of a channel respectively.
[0035] In this embodiment, see Figure 2 As shown, the SENet module includes a Squeeze module 、Excitation module With the Scale module First, after the data U is input into the SENet module, it passes through the compression module The module converts each feature map into a standard quantity through a global average pooling operation. The converted data is The matrix , improve the global receptive field of the data, and its calculation formula is shown in formula (1): (1) Where: is the first channels, is the characteristic data of the i-th row and j-th column in the channel; After completing the compression task, in the incentive module First, the matrix is transformed into A dimensionality reduction operation is performed to convert the original dimension from C to C / r. After that, it is activated by a ReLU activation function and then restored to its original dimension by a fully connected layer. Finally, the weights of all connected layers are calculated by the Sigmoid function, and the correlation between channels is learned by using the weights. Module output The calculation formula is shown in formula (2): (2) Where: Sigmoid and ReLU represent two activation functions, W1 and W2 represent two convolutional layers, The Scale module in the SENet module performs a weighted operation, which multiplies the stimulated weights by their corresponding features, redistributes the feature weights, and updates the feature map, thereby enhancing important features and suppressing unimportant features.
[0036] In this embodiment, as shown, the feature pyramid structure in the ResNet101 network is integrated with an attention mechanism. ResNet101 uses a fixed ratio of length and width compression. Here, feature graphs C2, C3, C4, and C5 obtained by compressing the length and width twice, three times, four times, and five times are taken out, and the attention mechanism is added before the C1-C5 convolutional layers. The construction of the feature pyramid is as follows Figure 5 shown.
[0037] In this embodiment, the output of spatial data according to the relative position of the steel bars in the image in step (3) is as follows: Figure 6 As shown, the coordinate points of the extracted vertices are substituted into the calculation formula of the steel bar spacing (L1) and the steel bar cover thickness (H1). The calculation formula is as follows: (3) (4) Where: H is the total detection depth, L is the detection length, H y is the width of the GPR image, L x is the length of the GPR image, x and y are the relative coordinates of the central vertex of the hyperbola in the GPR image, and the spatial data of the spacing between the steel bars and the thickness of the protective layer are obtained and stored in Excel.
[0038] In step (3), the least squares method is used to perform curve fitting on the steel bar mask segmented from Mask RCNN, and the coordinate points of the vertices are extracted, such as Figure 7 As shown, the output x and y are respectively substituted into formulas (3) and (4) to obtain the spatial data including the spacing of the steel bars and the thickness of the protective layer, and the data are stored in Excel, in the order of "Wall-Transverse rebar", "Wall-Longitudinal rebar", "Plank-Transverse rebar", "Plank-Longitudinal rebar", "Parastate-Transverse rebar" and "Parastate-Longitudinal rebar", and such files are stored in a fixed location.
[0039] Use the "File Path" editing node in Dynamo to select the Excel file that stores the steel bar position information of the corresponding component, complete the preliminary import of the steel bar parameter information, and then read the data from the Excel table through the "Data.lmportExcel" editing node. When extracting and saving the steel bar parameters, the spacing and protective layer thickness of the steel bars are stored in the first and second columns of the corresponding Excel file respectively; at the same time, the editing node "List.GetltemAtindex" returns the data of the specified index column of the input table, and uses the "List.GetltemAtindex" editing node to read the data of the specified index column and generate a new list and output it, so as to extract and output the steel bar spacing and protective layer thickness information in two different lists, set as S and list P.
[0040] First, use the above parameters to construct symmetrical points on the upper and lower surfaces of the wall, then connect the corresponding points inside the structure to form a straight line, and finally give the straight steel bar entity and attributes and interact with the concrete model to form a three-dimensional model of the entire steel-concrete structure. The modeling process is as follows: Figure 8 As shown in abcd.
[0041] The coordinate point at the lower right corner of the structure can be output through the "Suface.PointAtParameter" editing node, and then the specific values of the point on the X, Y, and Z coordinate axes can be obtained using "Point.X", "Point.Y", and "Point.Z" respectively; Figure 8Taking the data in as an example, the coordinate values of the lower right corner of the wall are output as x=-6.782, y=-2.379, and z=0.400; then input the list of steel bar spacing, protective layer thickness, etc. output by the "List.GetltemAtindex" edit node into "List.Map", and create a "Formula" edit node at the other input end of "List.Map" to calculate the values in the list representing steel bar spacing, protective layer thickness, etc. with the corresponding x, y, and z to obtain a new list, which stores the coordinate values of a certain end point of the steel bar on the X, Y, and Z coordinate axes; by Figure 8 Take the middle longitudinal reinforcement as an example, set any end of the reinforcement It is easy to know that the longitudinal reinforcement of the wall is perpendicular to the (x, y) plane, so the x and y values of its two end points are the same, and z can be calculated based on the end points on the upper and lower bottom surfaces of the structure, that is; (5) (6) or (7) In the above formula: x, y, z are the coordinates of the horizontal axis, vertical axis and vertical axis of the wall reference point respectively, P i is the protective layer thickness of a certain wall longitudinal reinforcement, S i is the spacing of the longitudinal reinforcement of a wall, and h is the height of the wall. For the transverse reinforcement in the wall, the above principle can also be used. It is only necessary to determine the plane perpendicular to the reinforcement and then deduce the calculation formula. Similarly, the endpoints of the reinforcement inside the floor slab can also be parametrically created in the same way.
[0042] In summary, the coordinate values of the two ends of the steel bar can be obtained, and then the two points can be connected to generate a straight line through the "Line.ByStartPointEndPoint" editing node, such as Figure 8 As shown in (c), the output result of the node "Line.ByStartPointEndPoint" is then input into the "Rebar.FreeFormRebarByCurves" editing node to determine the spatial properties of the steel bar. At the same time, the size and type of the steel bar are selected using "RebarBarType" and input into the "Rebar.FreeFormRebarByCurves" editing node as the output result to determine other properties of the steel bar. Finally, run the parametric modeling result of Dynamo, create the three-dimensional model of the steel bar into the opened model file of the built concrete structure, complete the combination of the two, and realize the creation of a complete model of the steel-concrete structure, as shown in Figure 1. Fig. 9 shown.
Claims
1. An intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning. It is characterized in that The steps include: (1) Real-time data collection: After the main construction facilities of the subway station are completed, the steel bar radar wave data is collected using ground penetrating radar; (2) Data preprocessing: A denoising algorithm is used to remove the interference noise of the direct wave noise and the ground reflection wave in the collected steel bar radar wave data. The corresponding images are extracted into the training set and test set files through the program. The extracted corresponding image contents include: the original image of the ground penetrating radar data, labels, masks, and types; at the same time, a file path list is generated to provide path guidance for the subsequent extraction of feature data; (3) Steel bar positioning detection: OpenCV2 is used to read and write ground penetrating radar images, and the input image information is sent to the MaskRCNN model to segment and extract the hyperbolic features of the steel bar target. The least squares method is used for curve fitting to obtain the hyperbola and extract the coordinate points of the vertices. The spatial data is output according to the relative position of the steel bar in the image. (4) Classify and store steel bar data: According to the different characteristics of steel bar protective layer thickness in different structures, determine the steel bar radar wave data corresponding to the plate, column and wall respectively, generate and save them in the corresponding data files; use ground penetrating radar images to generate time intervals, determine the transverse and longitudinal bars, generate and save them in the corresponding data files; (5) Separation of digital and model: The Dynamo parametric tool is used to perform three-dimensional geometric projection of the steel bar spatial data in the data file. The steel bar spatial data is combined with BIM technology to reversely reconstruct the steel bars in the steel-concrete structure in the subway project. The steel bar spatial data is mapped to the model to form a visual GPR to BIM model, realizing the sharing of the overall health monitoring information of the in-service building components.
2. According to the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that In step (3), the Mask RCNN model is used to segment and extract the hyperbolic features of the steel bar target. Specifically, the algorithm with the Mask RCNN network as the core and the SENet module as the backbone network is used to detect the steel bar target.
3. According to the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that The output of spatial data according to the relative position of the steel bars in the image in step (3) is as follows: the coordinate points of the extracted vertices are inserted into the steel bar spacing L 1 The thickness of the steel bar cover H 1 The calculation formula of is as follows: (3) (4) Where: H is the total detection depth, L is the detection length, H y is the width of the GPR image, L x is the length of the GPR image, x and y are the relative coordinates of the central vertex of the hyperbola in the GPR image, and the spatial data of the spacing between steel bars and the thickness of the protective layer are obtained and stored in Excel.
4. According to the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that In step (3), the least squares method is used to curve fit the steel mask segmented from Mask RCNN, and the coordinate points of the vertices are extracted. The output x and y are substituted into the formula (3) and (4) Calculate the spatial data including the spacing between steel bars and the thickness of the protective layer.
5. According to the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that In step (4), the time interval of the ground penetrating radar image is used to determine the transverse reinforcement and the longitudinal reinforcement, and the generated and saved data are stored in the corresponding Excel file. Specifically, the time interval of the ground penetrating radar image is superimposed on the generated time interval of the ground penetrating radar image. When the superimposed value of the superimposed value is within 5, it is the data of the longitudinal reinforcement, and its spatial data is generated and saved in the Excel file of the longitudinal reinforcement of the corresponding structure; if the superimposed value is greater than 5, it is the data of the transverse reinforcement, and its superimposed value is reduced by 5 and saved in the Excel file of the transverse reinforcement of the corresponding structure.
6. According to the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that The specific steps of the reverse reconstruction of the steel bars in step (5) are as follows: import the steel bar data from Excel into Dynamo, classify and store the spacing and protective layer thickness of the steel bars, extract and output the spacing and protective layer thickness information of the steel bars into two different lists, set as list S and list P, use the parameters in the list to construct symmetrical points on the upper and lower surfaces of the wall, then connect the corresponding points inside the structure to form a straight line, and finally give the straight steel bar entities and attributes and interact with the concrete model to form a three-dimensional model of the entire steel-concrete structure.
7. According to the intelligent positioning method for hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that The program in step (2) is a Python program.
8. According to the method for intelligent positioning of hidden steel bar engineering based on ground penetrating radar and machine learning in claim 1, It is characterized in that The data files in steps (4) and (5) are Excel files.
Citation Information
Cited By
Hidden project analysis and recognition system based on artificial intelligence
CN120472241A
Subway ballastless track sleeper rapid replacement method adopting directional micro-damage splitting
CN121827158A
