A method for identifying karst caves based on shield slag detection

Abstract

The present application relates to a method for identifying karst caves based on shield slag detection, comprising the following steps: S1, determining the karst development section during shield tunneling according to geological survey results; S2, collecting shield tunneling slag images in the karst development section, making a target recognition data set for karst cave fillings, training images with karst cave fillings based on the YOLO model, and obtaining corresponding model weights; S3, within the range of the connecting bridge, the slag outlet of the shield machine screw conveyor is equipped with a slag detection camera and a slag belt weighing system; S4, when the shield cutter head torque and thrust fluctuate, and the amplitude of the fluctuation exceeds a preset threshold, the slag detection camera is called based on the model weight to perform slag image detection and slag weight detection; S5, the karst cave category is identified and the karst cave area on the excavation surface is calculated by the slag image detection results and slag weight. The method has low cost and fast recognition speed, and can effectively guide construction.

Description

Technical Field

[0001] The present application relates to the technical field of cave identification, and in particular to a cave identification method based on shield slag detection. Background Art

[0002] When constructing shield tunnels in karst strata, shield tunneling construction often encounters karst caves. Due to the concealment and uncertainty of karst caves, it is very easy to cause engineering accidents such as mud and water gushing from the excavation face, lining damage, and shield head crash, resulting in huge economic losses and casualties. At present, before shield construction in karst strata, further karst exploration and advance treatment are usually carried out. However, the karst development is complex, and the existing karst exploration and treatment technology is difficult to cover the karst caves in the site well, resulting in a large number of untreated karst caves encountered during shield tunneling, which brings risks to tunneling construction.

[0003] At present, among the technologies for identifying caves during shield tunneling, some are carried out through advance drilling of excavation faces, advance geological prediction and other means to carry out advance cave identification. However, advance drilling of excavation faces requires machine downtime, which affects the efficiency of shield tunneling construction. The application of advanced geological prediction technology in shield machines is not yet mature and has poor reliability. Summary of the invention

[0004] The purpose of this application is to provide a method for identifying caves based on shield slag detection. Through the image and quality detection of slag, a method for identifying caves is proposed to provide timely warning for shield excavation construction. This method has low cost and fast recognition speed, and can effectively guide construction.

[0005] In order to achieve the above object, the present invention provides a method for identifying a karst cave based on shield slag detection, comprising the following steps:

[0006] S1, according to the geological survey results, determine the karst development section during shield tunneling;

[0007] S2, collect shield tunneling and slag discharge images in the karst development section, make a karst cave filling target recognition dataset, train the images with karst cave filling based on the YOLO model, and obtain the corresponding model weights;

[0008] S3, within the range of the connecting bridge, the slag outlet of the shield machine screw conveyor is equipped with a slag detection camera and a slag belt weighing system;

[0009] S4: When the shield cutter head torque and thrust fluctuate and the amplitude exceeds the preset threshold, the slag detection camera is called based on the model weight to perform slag image detection and slag weight detection;

[0010] S5, identifying the type of the cave and calculating the area of ​​the cave on the excavation surface through the slag image detection results and the slag weight.

[0011] In an optional embodiment, in step S1, the geological survey needs to determine the type of strata, the distribution of strata, the physical and mechanical parameters of strata, and determine the scope of cave development, the type of caves and the physical and mechanical parameters of cave fillings in the tunnel site;

[0012] It also includes obtaining the density of the stratum, the type of cave fillings, and the density of the cave fillings.

[0013] In an optional embodiment, in step S2, the slag discharge image needs to include the slag discharge port of the screw conveyor and the slag on the slag discharge belt located below the slag discharge port;

[0014] The preparation of cave filling target recognition dataset includes: dataset enhancement, dataset annotation, dataset division, and dataset conversion.

[0015] In an optional embodiment, in step S2, data set enhancement is used to process the slag discharge image and the targets in the image to enrich the number of samples during training and improve the sample quality, including: using tiling targets to increase the number of cave filling targets in the image, using median filtering to improve the edge contour clarity of the cave filling targets, and using Mosaic enhancement to enrich the features of the cave filling targets;

[0016] The dataset annotation includes pixel-level annotation of the cave filling targets in the slag images, using points to determine the rough edges, and then connecting all the points to achieve detailed contour delineation;

[0017] Dataset division: The image dataset is divided into training set, validation set and test set. The training set is used for the training stage of the YOLO model, the validation set is used for the evaluation stage of the YOLO model, and the test set is used for the testing stage of the YOLO model for slag images.

[0018] Dataset conversion is used to convert the file format of the labeled dataset into the format required for YOLO model training.

[0019] In an optional embodiment, in step S2, deep learning training is iteratively performed based on the YOLO model by gradient descent and back propagation methods, and the loss function is minimized by fine-tuning the weights in the neurons. The loss function CIoU Loss The calculation formula includes:

[0020]

[0021] Where: B represents the area of ​​the prediction box, B i Represents the area of ​​the real box, Distance2 is the distance between the center point of the predicted box and the target box, Distance c is the length of the diagonal of the minimum circumscribed rectangle, and are the aspect ratios of the predicted box and the target box, respectively.

[0022] In an optional implementation, the YOLO model is used to call the slag detection camera, which can detect the type of slag in real time. If cave fillings appear in the slag, the cave fillings can be automatically tracked and marked, and displayed and stored in the host computer.

[0023] In an optional embodiment, the slag detection camera includes a high frame rate industrial camera, and a fill light device is provided on the connecting bridge for filling light to the target area. The slag detection camera is installed on the connecting bridge via a shock-absorbing pad.

[0024] In an optional embodiment, the slag belt weighing system includes a weighing bridge with a built-in weighing sensor, a slag belt speed sensor and a weighing instrument. The slag belt weighing system is used to detect the slag amount in real time. If a karst cave is encountered, the reduction of the slag amount can be sensed in time, and the output weighing data can be used to calculate the karst cave area on the excavation surface.

[0025] The calculation method of the accumulated muck weight Q is:

[0026]

[0027] Where: T is time; Q is the cumulative weight of the slag within T time; Rate (t) is the weight of the slag at a certain moment, and its calculation formula is:

[0028]

[0029] Where: Y is the proportionality coefficient, which is a dimensionless constant obtained through calibration; v(t) is the instantaneous speed of the slag conveyor at a certain moment; q(t) is the instantaneous weight of the slag per unit length of the slag conveyor at a certain moment;

[0030] The slag conveyor weighing system consists of two sets. The two sets of slag conveyor weighing systems are calibrated using the same calibration method and are cross-checked.

[0031] In an optional embodiment, in step S4, when the shield cutter head torque and thrust fluctuate, and the fluctuation amplitude is greater than 20%, a 1-minute slag image detection and slag quality detection are started, and the shield excavation distance, shield over-excavation size, and water injection weight within 1 minute are recorded;

[0032] The slag image detection is completed by calling the camera through the detection module in the YOLO target detector, and the trained YOLO model weights are used in the detection module; the detection and annotation results of the cave filling are displayed and stored in the host computer;

[0033] The slag image detection steps include:

[0034] S41, preprocessing of collected images: preprocessing the images, including locating the target of cave fillings, image enhancement and image denoising;

[0035] S42, image feature extraction: taking the preprocessed cave filling image as an input image, and performing feature extraction in a neural network;

[0036] S43, prediction of target bounding box: obtaining label data in the image by conversion, and predicting the target bounding box;

[0037] S44, loss regression: calculation of loss function;

[0038] The slag quality test is completed by the slag belt weighing system, which obtains the cumulative weight of the belt slag within 1 minute;

[0039] The shield machine's excavation distance, shield over-excavation size, and water injection weight within 1 minute will be displayed and recorded in the shield machine's host computer. During actual operation, this information can be exported and used through the shield machine's host computer.

[0040] In an optional embodiment, in step S5, if the output image of the slag detection camera marks the karst cave filling slag, it proves that a filled karst cave is encountered. The karst cave area on the excavation surface can be obtained by the cumulative weight of the belt slag within 1 minute and the theoretical slag mass of the shield tunneling without encountering a karst cave for 1 minute. The calculation formula is as follows:

[0041]

[0042] Where: S c is the cave area on the excavation surface, G t G is the theoretical slag quality of the shield tunneling without encountering a karst cave for 1 min. r is the accumulated weight of the slag belt weighing machine for 1 minute, ρ f is the density of the cave filling, L1 is the tunneling length of the shield machine in 1 min;

[0043] Among them, the theoretical slag quality G of the shield tunneling without encountering a karst cave for 1 minute t The calculation formula is:

[0044]

[0045] Where: D c is the outer diameter of the cutterhead, δ is the shield over-excavation size, ρ s is the density of the excavated stratum, G w The weight of water injected during 1 minute of shield tunneling;

[0046] If the output image of the slag detection camera does not mark the slag filling of the cave, but the weighing mass of the slag belt is significantly reduced, it proves that a cavity or a cave with less karst filling is encountered. The volume of the cave can be obtained by the cumulative weight of the belt slag within 1 minute and the theoretical slag mass of the shield tunneling without encountering a cave for 1 minute. The calculation formula is as follows:

[0047] .

[0048] When the shield encounters a cave during excavation, the slag will have obvious characteristics. The present invention combines the slag weighing mass and slag image detection to track and identify the cave fillings on the slag belt and feedback the cave area on the shield excavation surface, providing timely warning for shield excavation construction. This method has low cost and fast recognition speed, and can effectively guide construction.

[0049] Other features and advantages of the present application will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0051] Figure 1 It is a schematic diagram of the structure of the karst cave identification system based on shield slag detection in this application;

[0052] Figure 2 It is a schematic diagram of the process of detecting the slag discharge image;

[0053] Figure 3 This is a schematic diagram of the structure of the slag conveyor weighing system. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical scheme and advantages of the embodiments of the present application clearer, the technical scheme in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. The components of the embodiments of the present application described and shown in the drawings here can be arranged and designed in various different configurations.

[0055] In the description of this application, it should be noted that the terms "inside", "outside", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the application is usually placed when in use, which is only for the convenience of describing this application and simplifying the description, and does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on this application. In addition, the terms "first", "second", etc. are only used to distinguish the description, and cannot be understood as indicating or implying relative importance.

[0056] In the description of this application, it should also be noted that, unless otherwise clearly specified and limited, the terms "disposed" and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances.

[0057] Combination Figure 1 The structure of the cave identification system based on shield slag detection is shown in Figure 2-Figure 3 The karst cave identification method based on shield slag detection in this application is mainly used for karst cave identification during shield tunneling construction.

[0058] When constructing shield tunnels in karst-developed strata, shield excavation construction often encounters caves. The geological structure of caves often appears in limestone strata. There are many types of caves, such as empty caves, semi-filled caves and fully-filled caves. The fillings include soft clay, gravel and so on.

[0059] The karst caves in the strata have a great impact on tunnel engineering, including the construction period and the operation period. This application is mainly used in the construction period of tunnel engineering. The impact of karst caves on shield tunnel construction is mainly concentrated in two parts. One part is that during the shield excavation process, the shield machine encounters karst caves, which may cause water inrush, tool damage, excavation surface collapse, etc. The other part is that after the shield tunnel segment lining is assembled, the karst cave part will cause uneven force on the segment, which may cause segment damage or water seepage.

[0060] At present, special surveys of caves are generally carried out before engineering construction. However, it is difficult to detect caves clearly and thoroughly with current survey technology. It is often the case that the size of the detected caves does not match the actual size, or some caves are not detected, making it difficult to deal with during the shield excavation process.

[0061] The existing technologies for cave exploration during shield tunneling include advanced geological prediction systems and advanced drilling surveys. However, the reliability of the advanced geological prediction system is currently difficult to guarantee, and the advanced drilling survey requires shutdown, which affects construction efficiency.

[0062] At the same time, the existing technology also has the following problems: drilling holes on the shield excavation face and using the tube wave method to detect the distribution of caves within a certain depth range. However, the shield needs to be shut down during the drilling detection process, which affects the construction efficiency. The range of the tube wave method to detect caves is small and the cost is high.

[0063] In addition, the shield machine is equipped with an acoustic wave detection system to identify the caves on the back of the tunnel segments, but it is unable to detect the caves encountered on the shield excavation face. During the construction process, the shield machine generates a lot of noise that affects the detection results.

[0064] In this regard, this application proposes a method for identifying karst caves based on shield slag detection. If a karst cave is encountered during shield excavation, the type of karst cave encountered can be identified after the shield slag is discharged, and the size of the karst cave encountered on the excavation face can be obtained. After obtaining relevant information about the karst cave, targeted measures can be taken during shield construction.

[0065] The principles of the cave identification method in this application mainly include slag detection and the principle of cave identification angle.

[0066] The starting point of the slag detection is that during the shield tunneling process, the cutter head will cut the rock and soil in front into the soil chamber, and the lower part of the soil chamber will be connected to the screw conveyor to discharge the slag in the soil chamber onto the slag conveyor. The soil chamber and the screw conveyor are closed and pressurized, making it difficult to detect. Therefore, the slag on the slag conveyor is mainly detected.

[0067] Through slag detection, the type of karst cave encountered on the shield excavation face can be sensed in real time, and the approximate size of the karst cave encountered on the excavation face can be calculated.

[0068] Slag detection is mainly divided into two parts, one of which involves the slag image detection of the training model, and the other part involves the slag quality detection.

[0069] In the slag image detection part, during the shield slag discharge process, if it is a uniform rock layer, the slag discharged is also uniform rock slag. If it encounters a cave filling, clay or gravel that is obviously different from ordinary rock slag will appear on the belt, which can prove that a filled cave or a semi-filled cave with more fillings has been encountered.

[0070] Slag image detection is based on the YOLO model. By collecting images of slag on the slag belt for training, a model for identifying slag anomalies can be obtained. Then, based on the trained model, the detection camera can be called to track and identify the cave fillings on the slag belt.

[0071] The quality inspection of the slag is to inspect the weight of the slag transported on the belt, and record the instantaneous flow value and the cumulative excavation weight value. If the shield encounters a large cave or multiple caves, the weight difference of the slag produced by the shield after 1 minute of excavation will be obvious. This weight difference can be used to calculate the surface area of ​​the cave on the excavation surface.

[0072] It should be noted that the shield machine host computer collects and outputs information once every 1 minute, and the excavation speed is also controlled by mm / min. Therefore, the slag output in 1 minute is used for detection in this application. The excavation speed in karst-developed strata generally does not exceed 30 mm / min.

[0073] The starting point for the identification of karst caves is to combine the simultaneous detection of slag image and slag quality. If the shield cutter head torque and thrust fluctuate during shield excavation, and the fluctuation amplitude is greater than 20%, the shield cutter head may encounter a karst cave. By starting the two tests simultaneously, the type of karst cave and the area of ​​karst cave encountered on the shield excavation surface can be identified:

[0074] If the camera tracks the filling soil of the cave, it proves that a filled cave is encountered. The area of ​​the cave on the excavation surface can be obtained by weighing the belt weight of the soil discharged after 1 minute of excavation and the theoretical discharge weight of the soil discharged after 1 minute of excavation without encountering a cave. The calculation formula is as follows:

[0075]

[0076] Where: S c is the cave area on the excavation surface, G t G is the theoretical slag quality of the shield tunneling without encountering a karst cave for 1 min. r is the actual weight of the soil during shield tunneling for 1 min, ρ f is the density of the cave filling, and L1 is the tunneling length of the shield machine in 1 min.

[0077] Among them, the theoretical slag mass G of the shield tunneling without encountering a karst cave for 1 minute is t The calculation formula is:

[0078]

[0079] Where: D c is the outer diameter of the cutterhead, δ is the shield over-excavation size, ρ s is the density of the excavated stratum, G w It is the weight of water injected during 1 minute of shield tunneling.

[0080] (2) If the camera does not track the slag of the cave filling, but the weight of the belt discharge is significantly reduced, it proves that a void or a cave with less karst filling is encountered. The area of ​​the cave on the excavation surface can be obtained by weighing the belt mass of the slag after 1 minute of excavation and the theoretical discharge mass of the shield after 1 minute of excavation without encountering a cave. The calculation formula is as follows:

[0081] .

[0082] In the specific implementation process, the cave identification method based on shield slag detection in the present application includes the following steps:

[0083] S1, according to the geological survey results, determine the karst development section during shield tunneling;

[0084] S2, collect shield tunneling and slag discharge images in the karst development section, make a karst cave filling target recognition dataset, train the images with karst cave filling based on the YOLO model, and obtain the corresponding model weights;

[0085] S3, within the range of the connecting bridge, the slag outlet of the shield machine screw conveyor is equipped with a slag detection camera and a slag belt weighing system;

[0086] S4: When the shield cutter head torque and thrust fluctuate and the amplitude exceeds the preset threshold, the slag detection camera is called based on the model weight to perform slag image detection and slag weight detection;

[0087] S5, identifying the type of the cave and calculating the area of ​​the cave on the excavation surface through the slag image detection results and the slag weight.

[0088] From the perspective of different steps, in step S1, geological survey needs to determine the type of strata, the distribution of strata, the physical and mechanical parameters of strata, and determine the scope of cave development, the type of caves and the physical and mechanical parameters of cave fillings in the tunnel site; it also includes obtaining the density of the strata, the type of cave fillings and the density of the cave fillings.

[0089] Specifically, according to the geological survey results, the karst development section during shield tunneling is determined, and the density of the stratum ρ is obtained. s , the type of cave filling, and the density of the cave filling ρ f ; Collect shield tunneling and slag discharge images in the karst development section to create a karst cave filling target recognition dataset, and train images with karst cave fillings based on the YOLO model to obtain the corresponding model weights.

[0090] The geological survey must determine the type of strata, the distribution of strata, the physical and mechanical parameters of strata, as well as the scope of cave development, the type of caves and the physical and mechanical parameters of cave fillings within the tunnel site.

[0091] In step S2, the slag discharge image must contain the slag discharge port of the quasi-screw conveyor and the slag on the lower slag discharge belt.

[0092] The preparation of cave filling target recognition dataset includes: dataset enhancement, dataset annotation, dataset division, and dataset conversion.

[0093] Among them, data set enhancement is used to process slag images and targets in the images to enrich the number of samples during training and improve the sample quality, including: using tiled targets to increase the number of cave filling targets in the image, using median filtering to improve the edge contour clarity of cave filling targets, and using Mosaic enhancement to enrich the features of cave filling targets.

[0094] Through data set enhancement, the training accuracy of the YOLO model can be effectively improved, thereby improving the detection and recognition accuracy of cave fillings.

[0095] The dataset annotation includes pixel-level annotation of the cave filling targets in the slag images, using points to determine the coarse edges, and then connecting all the points to achieve detailed contour outlining, in order to clarify the recognition targets in the YOLO model.

[0096] The data set is divided into training set, validation set and test set. The training set is used in the training stage of the YOLO model, the validation set is used in the evaluation stage of the YOLO model, and the test set is used in the testing stage of the YOLO model for slag images, that is, to complete the task of re-identifying the segmented images.

[0097] Dataset conversion is used to convert the file format of the labeled dataset into the format required for YOLO model training.

[0098] In step S2, the recognition is performed based on the YOLO model. The YOLO model is: using a deep neural network to identify and locate the target, which can be used to detect targets in complex scenes in real time. The network structure is divided into four parts: input end, backbone segment, neck segment and head output end. The model has a complete set of detection processes, including adjusting the input image size, extracting features through the backbone network, using the head part network fusion processing and outputting three layers of feature maps of different sizes, and finally outputting the prediction results. Compared with other deep learning models, the YOLO model has a fast reasoning speed and is suitable for real-time target detection applications. It can locate and detect in time during the rapid slag discharge process of the shield, and can detect general slag and cave fillings and give annotations.

[0099] During the training process, deep learning training is iterated through gradient descent and back propagation methods, and the loss function is minimized by fine-tuning the weights in the neurons. The loss function CIoU LossThe calculation formula includes:

[0100]

[0101] Where: B represents the area of ​​the prediction box, B i Represents the area of ​​the real box, Distance2 is the distance between the center point of the predicted box and the target box, Distance c is the length of the diagonal of the minimum circumscribed rectangle, and are the aspect ratios of the predicted box and the target box, respectively.

[0102] Within the scope of the connecting bridge, a slag detection camera and a slag belt weighing system are installed near the slag outlet of the shield machine's screw conveyor.

[0103] From the perspective of slag image detection, the YOLO model is used to call the slag detection camera, which can detect the type of slag in real time. If there is cave filling in the slag, it can automatically track and mark the cave filling, and display and store it in the host computer.

[0104] The function of the slag detection camera is to detect the type of slag at the slag outlet of the screw conveyor. The YOLO model is used to call the slag detection camera to detect the type of slag in real time. If there is karst filling in the slag, it can be automatically tracked and marked, and displayed and stored in the host computer. In this way, the karst filling in the discharged slag can be identified in real time without manual observation, and the type of karst cave encountered can be directly identified in the host computer.

[0105] In order to ensure the detection effect, the slag detection camera needs to use a high frame rate industrial camera. The slag outlet of the screw conveyor and the slag on the belt move quickly, and the frame rate of the industrial camera must be able to meet the high-speed movement of the slag.

[0106] During the detection process, a fill light device must be installed on the connecting bridge to fill in the light of the target area. If the environment is dim, it is difficult for the camera to ensure clear images during operation, and the detection results of the YOLO model will also be affected. In addition, the vibration at the connecting bridge is generally large, and the installation place of the slag detection camera must be shock-proof and installed on the connecting bridge with a shock-absorbing pad.

[0107] From the perspective of slag quality detection, the slag belt weighing system includes a weighing bridge with a built-in weighing sensor, a slag belt speed sensor and a weighing instrument. The slag belt weighing system is used to detect the slag amount in real time. If a cave is encountered, the reduction in the slag amount can be sensed in time, and the output weighing data can be used to calculate the cave area on the excavation surface.

[0108] Specifically, the weighing bridge transmits the gravity on the bottom roller of the belt to the weighing sensor, which is a device that converts force signals into analog electrical signals. The slag belt speed sensor outputs the electrical signal of the slag belt speed to the weighing meter. The frequency of the electrical signal is usually proportional to the running speed of the slag belt. The weighing meter integrates the measured value of the weighing sensor by detecting the period of the speed signal to obtain the accumulated weight of the slag within a certain period of time. This slag belt weighing system can detect the slag amount in real time. If a cave is encountered, the reduction of the slag amount can be sensed in time, and the output weighing data can be used to calculate the cave area on the excavation surface.

[0109] The calculation method of the accumulated muck weight Q is:

[0110]

[0111] Where: T is time; Q is the cumulative weight of the slag within T time; Rate (t) is the weight of the slag at a certain moment, and its calculation formula is:

[0112]

[0113] Where: Y is the proportionality coefficient, which is a dimensionless constant obtained through calibration; v(t) is the instantaneous speed of the slag conveyor at a certain moment; q(t) is the instantaneous weight of the slag per unit length of the slag conveyor at a certain moment;

[0114] The slag belt weighing system is installed near the spiral slag outlet, but not directly below the screw conveyor slag outlet, because there is a certain distance between the screw conveyor slag outlet and the slag belt, and the impact of the slag falling onto the belt will interfere with the weighing result. Therefore, it is installed in the range of the connecting bridge and close to the screw conveyor slag outlet. The water and fine particles in the slag are less lost within this range, and the weighing result is more accurate.

[0115] If the moisture content of shield slag is too high, water may intrude the contact surface between the slag belt speed sensor and the slag belt, causing the slag belt speed sensor to slip on the slag belt, resulting in the loss of weighing data. Therefore, two sets of slag belt weighing systems must be configured, and the distance between the two sets of slag belt weighing systems is controlled at about 5m. Use the same calibration method to calibrate the two weighing systems to ensure that the output weight values ​​of the two sets of slag belt weighing systems are consistent. The two sets of slag belt weighing systems can be checked against each other, greatly improving the reliability of slag weighing data. In field applications, if the slag weighing values ​​of the two sets of slag belt weighing systems are significantly different, the slag belts can be cleaned or the weighing systems can be recalibrated to assist on-site personnel in maintenance.

[0116] In step S4, when the shield cutter torque and thrust fluctuate and the fluctuation amplitude is greater than 20%, the slag image detection and slag quality detection for 1 minute are started, and the shield excavation distance L1, shield over-excavation size δ, and water injection weight G within 1 minute are recorded. w .

[0117] When the shield machine encounters a cave during excavation, the excavation surface in front of the cutterhead will be unevenly soft and hard. During the rotation of the cutterhead at the rated speed, the torque will change significantly, with a change range of not less than 20%, and the cutterhead thrust will also fluctuate significantly. This abnormal parameter jump proves that a cave has been encountered.

[0118] The slag image detection is completed by calling the camera through the detection module in the YOLO target detector. The trained YOLO model weights are used in the detection module. The detection and annotation results of the cave fillings are displayed and stored in the host computer.

[0119] The slag image detection steps include:

[0120] S41, preprocessing of collected images: preprocessing the images, including locating the target of cave fillings, image enhancement and image denoising;

[0121] S42, image feature extraction: taking the preprocessed cave filling image as an input image, and performing feature extraction in a neural network;

[0122] S43, prediction of target bounding box: obtaining label data in the image by conversion, and predicting the target bounding box;

[0123] S44, loss regression: calculation of loss function;

[0124] The slag quality test is completed by the slag belt weighing system, and the cumulative weight of the belt slag within 1 minute is obtained. r ;

[0125] The excavation distance L1 of the shield within 1 minute, the shield over-excavation size δ, and the water injection weight G w , this information will be displayed and recorded in the host computer of the shield machine, and can be exported and used through the host computer of the shield machine during actual operation.

[0126] In step S5, the angle of identification of the cave type and cave size is identified and calculated by the host computer to identify the cave type (filled cave or hollow cave) and obtain the cave area on the excavation surface.

[0127] If the output image of the slag detection camera shows the slag filling, it means that a filled karst cave is encountered. The karst cave area on the excavation surface can be obtained by the cumulative weighing mass of the slag belt scale for 1 minute and the theoretical slag mass of the shield tunneling without encountering a karst cave for 1 minute. The calculation formula is as follows:

[0128]

[0129] Where: S c is the cave area on the excavation surface, G t G is the theoretical slag quality of the shield tunneling without encountering a karst cave for 1 min. r is the actual weight of the soil during shield tunneling for 1 min, ρ f is the density of the cave filling, and L1 is the tunneling length of the shield machine in 1 min.

[0130] Among them, the theoretical slag mass G of the shield tunneling without encountering a karst cave for 1 minute is t The calculation formula is:

[0131]

[0132] Where: D c is the outer diameter of the cutterhead, δ is the shield over-excavation size, ρ s is the density of the excavated stratum, G w It is the weight of water injected during 1 minute of shield tunneling.

[0133] If the output image of the slag detection camera does not mark the slag filling soil, but the mass weighed by the slag belt is significantly reduced, it proves that a cavity or a karst cave with less karst filling is encountered. The volume of the karst cave can be obtained by the cumulative weight of the slag belt scale for 1 minute and the theoretical slag mass of the shield tunneling without encountering a karst cave for 1 minute. The calculation formula is as follows:

[0134] .

[0135] Through the cave identification method in this application, slag image detection and slag quality detection are effectively combined, which can greatly reduce the operational difficulty of existing cave identification methods, improve detection efficiency, and reduce detection costs, and can provide timely and accurate guidance for subsequent shield excavation construction.

[0136] It should be noted that, in the absence of conflict, the features in the embodiments of this application may be combined with each other.

[0137] The above description is only the preferred embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A method for identifying karst caves based on shield slag detection, characterized in that: The following steps are involved: S1, according to the geological survey results, determine the karst development section during shield tunneling; S2, collect shield tunneling and slag discharge images in the karst development section, make a karst cave filling target recognition dataset, train the images with karst cave filling based on the YOLO model, and obtain the corresponding model weights; S3, within the range of the connecting bridge, the slag outlet of the shield machine screw conveyor is equipped with a slag detection camera and a slag belt weighing system; The slag belt weighing system includes a weighing bridge with a built-in weighing sensor, a slag belt speed sensor and a weighing instrument. The slag belt weighing system is used to detect the slag amount in real time. If a karst cave is encountered, the reduction of the slag amount can be sensed in time, and the output weighing data can be used to calculate the karst cave area on the excavation surface. The calculation method of the accumulated muck weight Q is: Where: T is time; Q is the cumulative weight of the slag within T time; Rate (t) is the weight of the slag at a certain moment, and its calculation formula is: Where: Y is the proportionality coefficient, which is a dimensionless constant obtained through calibration; v(t) is the instantaneous speed of the slag conveyor at a certain moment; q(t) is the instantaneous weight of the slag per unit length of the slag conveyor at a certain moment; There are two sets of weighing systems for the slag conveyor belts. The two sets of weighing systems for the slag conveyor belts are calibrated using the same calibration method and are mutually calibrated. S4: When the shield cutter head torque and thrust fluctuate and the amplitude exceeds the preset threshold, the slag detection camera is called based on the model weight to perform slag image detection and slag weight detection; In step S4, when the shield cutter torque and thrust fluctuate, and the fluctuation amplitude is greater than 20%, 1-minute slag image detection and slag quality detection are started, and the shield excavation distance, shield over-excavation size, and water injection weight within 1 minute are recorded; The slag image detection is completed by calling the camera through the detection module in the YOLO target detector, and the trained YOLO model weights are used in the detection module; the detection and annotation results of the cave filling are displayed and stored in the host computer; The slag image detection steps include: S41, preprocessing of collected images: preprocessing the images, including locating the target of cave fillings, image enhancement and image denoising; S42, image feature extraction: taking the preprocessed cave filling image as an input image, and performing feature extraction in a neural network; S43, prediction of target bounding box: obtaining label data in the image by conversion, and predicting the target bounding box; S44, loss regression: calculation of loss function; The slag quality test is completed by the slag belt weighing system, which obtains the cumulative weight of the belt slag within 1 minute; The excavation distance of the shield machine within 1 minute, the shield over-excavation size, and the water injection weight will be displayed and recorded in the shield machine's host computer. In the actual operation process, this information can be exported and used through the shield machine's host computer; S5, identifying the type of karst cave and calculating the area of ​​karst cave on the excavation surface through the slag image detection result and the slag weight; In step S5, if the output image of the slag detection camera shows the slag filling in the cave, it means that a filled karst cave has been encountered. The karst cave area on the excavation surface can be obtained by the cumulative weight of the belt slag within 1 minute and the theoretical slag mass of the shield tunneling without encountering a karst cave for 1 minute. The calculation formula is as follows: Where: S c is the cave area on the excavation surface, G t G is the theoretical slag quality of the shield tunneling without encountering a karst cave for 1 min. r is the accumulated weight of the slag belt weighing machine for 1 minute, ρ f is the density of the cave filling, L1 is the tunneling length of the shield machine in 1 min; Among them, the theoretical slag quality G of the shield tunneling without encountering a karst cave for 1 minute t The calculation formula is: Where: D c is the outer diameter of the cutterhead, δ is the shield over-excavation size, ρ s is the density of the excavated stratum, G w The weight of water injected during 1 minute of shield tunneling; If the output image of the slag detection camera does not mark the slag filling of the cave, but the weighing mass of the slag belt is significantly reduced, it proves that a cavity or a cave with less karst filling is encountered. The volume of the cave can be obtained by the cumulative weight of the belt slag within 1 minute and the theoretical slag mass of the shield tunneling without encountering a cave for 1 minute. The calculation formula is as follows: 。 2. The method for identifying karst caves based on shield slag detection according to claim 1 is characterized in that: In step S1, the geological survey needs to determine the type of strata, the distribution of strata, the physical and mechanical parameters of strata, and determine the scope of cave development, the type of cave and the physical and mechanical parameters of cave fillings in the tunnel site; It also includes obtaining the density of the stratum, the type of cave fillings, and the density of the cave fillings.

3. The method for identifying karst caves based on shield slag detection according to claim 1 is characterized in that: In step S2, the slag discharge image needs to include the slag discharge port of the screw conveyor and the slag on the slag discharge belt located below the slag discharge port; The preparation of cave filling target recognition dataset includes: dataset enhancement, dataset annotation, dataset division, and dataset conversion.

4. The method for identifying karst caves based on shield slag detection according to claim 3 is characterized in that: In step S2, data set enhancement is used to process the slag images and the targets in the images to enrich the number of samples during training and improve the sample quality, including: using tiling targets to increase the number of cave filling targets in the image, using median filtering to improve the edge contour clarity of the cave filling targets, and using Mosaic enhancement to enrich the features of the cave filling targets; The dataset annotation includes pixel-level annotation of the cave filling targets in the slag images, using points to determine the rough edges, and then connecting all the points to achieve detailed contour delineation; Dataset division: The image dataset is divided into training set, validation set and test set. The training set is used for the training stage of the YOLO model, the validation set is used for the evaluation stage of the YOLO model, and the test set is used for the testing stage of the YOLO model for slag images. Dataset conversion is used to convert the file format of the labeled dataset into the format required for YOLO model training.

5. The method for identifying karst caves based on shield slag detection according to claim 3 is characterized in that: In step S2, deep learning training is iterated based on the YOLO model through gradient descent and back propagation methods, and the loss function is minimized by fine-tuning the weights in the neurons. The loss function CIoU Loss The calculation formula includes: Where: B represents the area of ​​the prediction box, B i Represents the area of ​​the real box, Distance2 is the distance between the center point of the predicted box and the target box, Distance c is the length of the diagonal of the minimum circumscribed rectangle, and are the aspect ratios of the predicted box and the target box, respectively.

6. The method for identifying karst caves based on shield slag detection according to claim 3 is characterized in that: The YOLO model is used to call the slag detection camera, which can detect the type of slag in real time. If there is cave filling in the slag, it can automatically track and mark the cave filling, and display and store it in the host computer.

7. The method for identifying karst caves based on shield slag detection according to claim 6 is characterized in that: The slag detection camera includes a high frame rate industrial camera. A fill light device is provided on the connecting bridge for filling light to the target area. The slag detection camera is installed on the connecting bridge through a shock-absorbing pad.

Images