Shield tunnel crack intelligent repairing method based on image processing and deep learning

Through image processing and deep learning technology, tunnel cracks are identified and positioned, and construction parameters are adjusted in combination with grouting volume, the problems of low efficiency and unstable quality repair of existing shield tunnel cracks are solved, intelligent repair is achieved, and construction safety is improved and effects.

CN119006356APending Publication Date: 2024-11-22CHENGDUN TUNNEL UNDERGROUND ENG CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202410662788.6
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-05-27
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

The repair of cracks in existing shield tunnels mainly relies on manual operations, with low efficiency and unstable quality, and the repair materials may pose a threat to the health of construction workers.

Method used

Using an intelligent repair method based on image processing and deep learning, the tunnel crack pictures are collected through drones, combined with an ultrasonicer to measure the crack depth, a deep learning model is constructed for crack identification and positioning, and the crack volume is calculated using an integrated CNN model, and the crack volume is employed and slurry is performed. Adjust construction parameters in quantity to achieve intelligent repair.

Benefits of technology

It improves the efficiency and quality of tunnel crack repair, reduces the health risks to construction workers, and ensures the reliability of the repair effect.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006356A_ABST
    Figure CN119006356A_ABST
Patent Text Reader

Abstract

An intelligent shield tunnel crack repairing method based on image processing and deep learning comprises the steps that an unmanned aerial vehicle carrying a high-speed camera is adopted to collect crack pictures in a tunnel, and an ultrasonic instrument is adopted to measure the depth of each crack; the collected pictures are selected and processed, cracks are marked, and the lengths and the widths of the cracks are determined through the image segmentation technology; taking the marked pictures as an original data set, and dividing the original data set into a training set, a verification set and a test set; constructing a deep learning recognition model, constructing an integrated CNN model, during tunnel crack repair, continuously shooting pictures in a tunnel by using a high-speed camera of a repair system, automatically recognizing and positioning the crack position by using the deep learning model, and calculating the crack depth by using the integrated CNN model; and the mechanical arm with the grouting repairing device is controlled to conduct grouting on the crack at the designated position till all crack repairing work is completed. The repairing quality and repairing efficiency of tunnel cracks are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the technical field of intelligent repair of cracks in shield tunnels, and relates to an intelligent repair method for cracks in shield tunnels based on image processing and deep learning. Background Art

[0002] With the rapid development of social economy, the ground space can no longer meet the development needs of large cities, especially megacities, and the development of underground space has become the only way to achieve sustainable urban development. Due to the large-scale construction of urban subways, shield tunnels are the main structural form for the development and utilization of underground space. During the operation of the tunnel, cracks in the tunnel lining are inevitable due to traffic loads and other factors. If they are not repaired in time, they may have a huge impact on the safe operation of the tunnel. At present, the repair of tunnel cracks still mainly adopts manual repair methods, which not only has high work intensity and low work efficiency, but also has uneven repair quality and cannot guarantee the repair effect. In addition, some repair materials themselves have certain toxicity, which will cause certain harm to the health of construction workers during the repair construction.

[0003] In recent years, deep learning technology has received increasing attention and has been widely used in many fields, such as image recognition, target detection, model regression, etc., and has good performance and application potential. At present, in the field of tunnel disease repair, deep learning technology is mainly used to detect superficial diseases such as tunnel leakage or defects. In order to improve the repair quality and efficiency of tunnel cracks, it is urgent to invent an efficient and intelligent tunnel crack intelligent repair method. Summary of the invention

[0004] In order to overcome the shortcomings of the existing technology, the present invention provides an intelligent repair method for shield tunnel cracks based on image processing and deep learning, which can not only quickly identify and locate cracks in the tunnel, but also judge the repair effect, thereby improving the repair efficiency while improving the repair quality.

[0005] The technical solution adopted by the present invention to solve its technical problem is:

[0006] An intelligent repair method for shield tunnel cracks based on image processing and deep learning, comprising the following steps:

[0007] (1) Use a drone equipped with a high-speed camera to collect images of cracks in the tunnel, and use an ultrasonic device to measure the depth of each crack;

[0008] (2) Select and process the collected images, remove distorted, blurred, low-resolution images, and images with edge cracks, and resize all images to the same pixel size m×n;

[0009] (3) Annotate all the cracks in the processed images and use image segmentation technology to determine the length and width of the cracks;

[0010] (4) Take the annotated images as the original data set, divide the original data set into three parts: training set, validation set and test set, and set the ratio;

[0011] (5) Construct a deep learning recognition model, including input layer, convolution layer, pooling layer, activation layer, and output layer. The input layer receives an m×n×1 image, that is, a grayscale image with m rows and n columns, and then reduces it to 1×1×64 through the convolution layer and pooling layer; the activation layer predicts the result according to the classification type, and finally the output layer outputs the classification prediction result and the crack location;

[0012] (6) Constructing an integrated CNN model, which includes a traditional CNN model and a regression model. The original data set is input into the CNN model to extract data features. The extracted data features and the corresponding crack depth are then input into the regression model as independent variables X and dependent variables Y, respectively. The relationship between the independent variable X and the dependent variable Y is established, and the crack depth D can be predicted based on the data set.

[0013] (7) When repairing tunnel cracks, the high-speed camera of the repair system is used to continuously take pictures of the tunnel. The trained deep learning model is used to identify and locate the crack position, and the image processing module obtains the length L and width W of the crack;

[0014] (8) The system uses the trained integrated CNN model to calculate the crack depth D, and calculates the volume of each crack V = L × W × D based on the calculated crack length L, width W and depth D;

[0015] (9) Control the robotic arm with the grouting repair device to the crack at the specified position for grouting. The initial grouting pressure is set to P0 and the maximum grouting pressure is set to P m , after grouting is completed, record the grouting volume Q:

[0016] When the grouting volume Q is less than 0.5V, the grouting pump is controlled by the repair system to maintain the grouting pressure to P0 and continue grouting;

[0017] When the grouting volume Q is between 0.5V and 0.8V, the grouting pump is controlled by the repair system to adjust the grouting pressure to (P0+P m ) / 2, continue grouting;

[0018] When the grouting volume Q is between 0.8V and 1.0V, the grouting pump is controlled by the repair system to adjust the grouting pressure to P m , continue grouting;

[0019] When the grouting volume Q is greater than or equal to V, grouting can be stopped and the next crack repair can be started.

[0020] (10) Carry out tunnel crack repair work according to steps 7 to 9 until all crack repair work is completed.

[0021] Furthermore, in step (6), the integrated CNN model includes two parts: a CNN model and a regression model. The CNN model is used to extract the data features of the cracks, and the regression model is used to establish the relationship between the data features and the crack depth D.

[0022] Preferably, the regression model may adopt a random forest model or an extreme gradient boosting XGBoost model to improve prediction accuracy.

[0023] The beneficial effects of the present invention are mainly manifested in: by adopting image processing and deep learning methods, a database is constructed with a large number of crack pictures in the tunnel, and a tunnel crack recognition, positioning and calculation model is constructed. By comparing the crack volume with the grouting amount, the repair effect is judged, and then the construction parameters are adjusted to realize the intelligent grouting repair construction, improve the repair efficiency, and ensure the repair quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a flow chart of the intelligent repair method for shield tunnel cracks based on image processing and deep learning. DETAILED DESCRIPTION

[0025] The present invention will be further described below in conjunction with the accompanying drawings.

[0026] Reference Figure 1 , an intelligent repair method for shield tunnel cracks based on image processing and deep learning, comprising the following steps:

[0027] (1) Use a drone equipped with a high-speed camera to collect images of cracks in the tunnel, and use an ultrasonic device to measure the depth of each crack;

[0028] (2) Select and process the collected images, remove distorted, blurred, low-resolution images, and images with edge cracks, and resize all images to the same pixel size m×n;

[0029] (3) Annotate all the cracks in the processed images and use image segmentation technology to determine the length and width of the cracks;

[0030] (4) Take the annotated images as the original data set, and divide the original data set into three parts: training set, validation set and test set, and set the ratio;

[0031] (5) Construct a deep learning recognition model, including input layer, convolution layer, pooling layer, activation layer, and output layer. The input layer receives an m×n×1 image, that is, a grayscale image with m rows and n columns, and then reduces it to 1×1×64 through the convolution layer and pooling layer; the activation layer predicts the result according to the classification type, and finally the output layer outputs the classification prediction result and the crack location;

[0032] (6) Constructing an integrated CNN model, which includes a traditional CNN model and a regression model. The original data set is input into the CNN model to extract data features. The extracted data features and the corresponding crack depth are then input into the regression model as independent variables X and dependent variables Y, respectively. The relationship between the independent variable X and the dependent variable Y is established, and the crack depth D can be predicted based on the data set.

[0033] In the step (6), the integrated CNN model includes two parts: a CNN model and a regression model. The CNN model is used to extract the data features of the cracks, and the regression model is used to establish the relationship between the data features and the crack depth D.

[0034] Preferably, the regression model may adopt a random forest model or an extreme gradient boosting XGBoost model to improve prediction accuracy;

[0035] (7) When repairing tunnel cracks, the high-speed camera of the repair system is used to continuously take pictures of the tunnel. The trained deep learning model is used to identify and locate the crack position, and the image processing module obtains the length L and width W of the crack;

[0036] (8) The system uses the trained integrated CNN model to calculate the crack depth D, and calculates the volume of each crack V = L × W × D based on the calculated crack length L, width W and depth D;

[0037] (9) Control the robotic arm with the grouting repair device to the crack at the specified position for grouting. The initial grouting pressure is set to P0 and the maximum grouting pressure is set to P m , after grouting is completed, record the grouting volume Q:

[0038] When the grouting volume Q is less than 0.5V, the grouting pump is controlled by the repair system to maintain the grouting pressure to P0 and continue grouting;

[0039] When the grouting volume Q is between 0.5V and 0.8V, the grouting pump is controlled by the repair system to adjust the grouting pressure to (P0+P m ) / 2, continue grouting;

[0040] When the grouting volume Q is between 0.8V and 1.0V, the grouting pump is controlled by the repair system to adjust the grouting pressure to P m , continue grouting;

[0041] When the grouting volume Q is greater than or equal to V, grouting can be stopped and the next crack repair can be started.

[0042] (10) Carry out tunnel crack repair work according to steps (7) to (9) until all crack repair work is completed.

[0043] Example: An intelligent repair method for shield tunnel cracks based on image processing and deep learning, including the following steps:

[0044] (1) A total of 5,000 images of cracks in the tunnel were collected using a drone equipped with a high-speed camera, and the depth Di of each crack was measured using an ultrasonic instrument;

[0045] (2) The collected images were manually selected and processed to remove distorted, blurred, low-resolution, and edge-cracked images, and all images were resized to 1280 × 1028 pixels;

[0046] (3) Manually mark the cracks in all processed images and use image segmentation technology to determine the length Li and width Wi of the cracks;

[0047] (4) The labeled images are used as the original data set, and the original data set is divided into three parts: training set, validation set and test set, with a ratio of 60%:20%:20%, that is, 3000 training sets, 1000 validation sets and 1000 test sets;

[0048] (5) Construct a deep learning recognition model, including input layer, convolution layer, pooling layer, activation layer and output layer. The input layer receives a 1280×1028×1 image, that is, a grayscale image with 1280 rows and 1028 columns. Then it is reduced to 1×1×64 through the convolution layer and pooling layer. The activation layer uses the Softmax function and the prediction result is finally output according to the classification type.

[0049] (6) Constructing an integrated CNN model, which includes a traditional CNN model and a regression model. The regression model uses Random Forest or extreme gradient boosting XGBoost. The original data set is input into the CNN model to extract data features. The extracted data features and the corresponding crack depth are then input into the regression model as independent variables X and dependent variables Y, respectively. The relationship between the independent variable X and the dependent variable Y is established, and the crack depth D can be predicted according to the data set.

[0050] The integrated CNN model includes a CNN model and a regression model. The CNN model is used to extract data features of the cracks, and the regression model is used to establish a relationship between the data features and the crack depth D. The regression model can use a random forest model and an extreme gradient enhancement XGBoost model to improve prediction accuracy.

[0051] (7) When repairing tunnel cracks, the high-speed camera of the repair system is used to continuously take pictures of the tunnel. The trained deep learning model is used to identify and locate the crack positions. The built-in image processing module obtains the length Li and width Wi of each crack.

[0052] (8) The trained integrated CNN model is used to calculate the depth Di of each crack, and based on the calculated length Li, width Wi and depth Di of each crack, the volume of each crack Vi = Li × Wi × Di can be calculated.

[0053] (9) Control the robotic arm with the grouting repair device to the crack at the specified position for grouting. The initial grouting pressure is set to P0 and the maximum grouting pressure is set to P m , after grouting is completed, record the grouting volume Q:

[0054] When the grouting volume Q is less than 0.5V, the grouting pump is controlled by the repair system to maintain the grouting pressure to P0 and continue grouting;

[0055] When the grouting volume Q is between 0.5V and 0.8V, the grouting pump is controlled by the repair system to adjust the grouting pressure to (P0+P m ) / 2, continue grouting;

[0056] When the grouting volume Q is between 0.8V and 1.0V, the grouting pump is controlled by the repair system to adjust the grouting pressure to P m , continue grouting;

[0057] When the grouting volume Q is greater than or equal to V, grouting can be stopped and the next crack repair can be started.

[0058] (10) Completing the tunnel crack repair work: Carry out the tunnel crack repair work according to steps (7) to (9) until all crack repair work is completed.

[0059] This embodiment uses a drone equipped with a high-speed camera to collect images of cracks in the tunnel, and uses ultrasound to measure the depth of each crack as the original data set. A deep learning model is constructed to intelligently identify and locate the crack position, calculate the crack size and volume, and judge the quality of crack repair by the grouting amount, and then adjust the construction parameters to ensure the repair effect.

[0060] The contents described in the embodiments of this specification are merely enumerations of implementation forms of the inventive concept and are for illustrative purposes only. The protection scope of the present invention should not be considered to be limited to the specific forms described in this embodiment, and the protection scope of the present invention also extends to equivalent technical means that can be thought of by ordinary technicians in this field based on the inventive concept.

Claims

1. An intelligent repair method for shield tunnel cracks based on image processing and deep learning, It is characterized in that The method comprises the following steps: (1) Use a drone equipped with a high-speed camera to collect images of cracks in the tunnel, and use an ultrasonic device to measure the depth of each crack; (2) Select and process the collected images, remove distorted, blurred, low-resolution images, and images with edge cracks, and resize all images to the same pixel size m×n; (3) Annotate all the cracks in the processed images and use image segmentation technology to determine the length and width of the cracks; (4) Take the annotated images as the original data set, and divide the original data set into three parts: training set, validation set and test set, and set the ratio; (5) Construct a deep learning recognition model, including an input layer, a convolution layer, a pooling layer, an activation layer, and an output layer. The input layer receives an m×n×1 image, that is, a grayscale image with m rows and n columns, and then reduces it to 1×1×64 through a convolution layer and a pooling layer; the activation layer predicts the result according to the classification type, and finally the output layer outputs the classification prediction result and the crack location; (6) Constructing an integrated CNN model, which includes a CNN model and a regression model. The original data set is input into the CNN model to extract data features. The extracted data features and the corresponding crack depth are then input into the regression model as independent variables X and dependent variables Y, respectively. The relationship between the independent variable X and the dependent variable Y is established, and the crack depth D can be predicted based on the data set. (7) When repairing tunnel cracks, the high-speed camera of the repair system is used to continuously take pictures of the tunnel. The trained deep learning model is used to identify and locate the crack position, and the image processing module obtains the length L and width W of the crack; (8) Using the trained integrated CNN model to calculate the crack depth D, and based on the calculated crack length L, width W, and depth D, calculate the volume of each crack V = L × W × D; (9) Control the robotic arm with the grouting repair device to the crack at the specified location for grouting. The initial grouting pressure is set to P 0 , the maximum grouting pressure is set to P m , after grouting is completed, record the grouting volume Q: When the grouting volume Q is less than 0.5V, the grouting pump is controlled by the repair system to maintain the grouting pressure to P 0 , continue grouting; When the grouting volume Q is between 0.5V and 0.8V, the grouting pump is controlled by the repair system to adjust the grouting pressure to (P 0 +P m ) / 2, continue grouting; When the grouting volume Q is between 0.8V and 1.0V, the grouting pump is controlled by the repair system to adjust the grouting pressure to P m , continue grouting; When the grouting volume Q is greater than or equal to V, the grouting can be stopped and the next crack repair can be started; (10) Carry out tunnel crack repair work according to steps 7 to 9 until all crack repair work is completed.

2. The shield tunnel crack intelligent repair method based on image processing and deep learning as claimed in claim 1, It is characterized in that In the step (6), the integrated CNN model includes two parts: a CNN model and a regression model. The CNN model is used to extract the data features of the cracks, and the regression model is used to establish the relationship between the data features and the crack depth D.

3. The shield tunnel crack intelligent repair method based on image processing and deep learning as claimed in claim 2, It is characterized in that The regression model adopts a random forest model or an extreme gradient boosting XGBoost model.

Citation Information

Patent Citations

  • Deep learning-based crack identification method

    CN107403197A

  • Deep learning-based rapid tunnel crack identification method

    CN108229461A

  • Segmented grouting method for tunnel large pipe shed

    CN114483091A

  • Tunnel crack automatic detection method based on deep learning

    CN116977857A

  • Deep learning-based tunnel crack rapid identification method

    CN117994710A