CN115047007B - Intelligent detection method for long-distance water diversion tunnel defects during operation - Google Patents

Abstract

The invention discloses an intelligent detection method for long-distance water diversion tunnel operation period diseases, including an AUV underwater robot and an ROV underwater robot, including the following steps: at the entrance of the water diversion tunnel, the ROV underwater robot deploys the AUV underwater robot; the AUV underwater robot performs acoustic 3D scanning on the entire tunnel, performs laser scanning on the entire tunnel again, and photographs the disease points scanned twice by a near optical detection component, and matches and compares the video frames with the disease database; if there is still an abnormal situation where the disease type cannot be determined after matching and comparison, the location of the abnormal situation is photographed and stored in the disease database; the ROV underwater robot carries the AUV underwater machine and is recovered by the control console. A rapid survey with acoustic detection as the main method and optical detection as the auxiliary method is adopted, and the disease points in the survey process are inspected in detail by close optical photography and laser ranging during the rapid survey.

Description

Intelligent detection method for long-distance water diversion tunnel defects during operation

Technical Field

The present invention relates to the field of internal disease detection of water diversion tunnels, and in particular to an intelligent disease detection method for long-distance water diversion tunnels during operation.

Background Art

The water diversion tunnel plays an important role in water conservancy projects. It is mainly used in water delivery and irrigation projects. However, long-term use will cause cracks and leakage in the tunnel wall. There are currently two detection methods, one is manual detection and the other is underwater robot detection. Manual detection is dangerous and costly, so this method has been basically eliminated. Underwater robots can be divided into two types, one is a cable underwater robot (ROV) and the other is an untethered underwater robot (AUV).

At present, the development of ROV is relatively mature. Because it has a cable, it has the advantages of convenient information transmission and sustainable energy supply. However, the cable also greatly limits its range of activities, making it difficult to cope with water diversion tunnels over 8 km. In addition, the ROV cable is easy to knot, break, and easily entangled in obstacles such as water plants, rocks, and sunken wood underwater, affecting the work of the ROV and even making it unable to return. The research on AUV started late and is relatively shallow. Because there is no constraint of the cable, it greatly increases its range of activities, and there is no problem of cable knotting and entanglement. However, AUV is constrained by energy and information transmission when working underwater.

Summary of the invention

In view of the above-mentioned defects in the prior art, a method for intelligent detection of diseases in long-distance water diversion tunnels during operation is provided, which not only improves the detection speed but also ensures the detection accuracy.

The technical solution adopted by the present invention to solve the above technical problems is:

1. An intelligent detection method for long-distance water diversion tunnel operation period diseases, comprising an AUV underwater robot for detection and an ROV underwater robot for providing energy replenishment and information exchange; characterized in that it comprises the following steps:

Step 1: Under the control of the operating console, the ROV underwater robot loads the AUV underwater robot and enters the entrance of the water diversion tunnel. At the entrance, the ROV underwater robot deploys the AUV underwater robot;

Step 2: The AUV underwater robot performs an acoustic 3D scan of the entire tunnel to obtain a complete 3D image of the diversion tunnel structure and determine the defect points, and save the first detection results;

Step 3: Perform laser scanning on the entire tunnel again to obtain complete high-density point cloud data of the entire tunnel wall and determine the defect points. Use the near-optical detection component to take close-up optical images of the defect points scanned twice, match and compare the video frames containing the defects with the defect database, and save and analyze the second detection results.

Step 4: If there are still abnormal conditions that cannot be determined whether there is a disease after two scans, the AUV underwater robot will take a video of the location of the abnormal condition and store it in the disease database. If there is no abnormal condition that cannot be determined whether there is a disease, it will directly proceed to step 5;

Step 5: The AUV underwater robot returns to the ROV underwater robot, and the ROV underwater robot carries the AUV underwater machine and is recovered by the console.

According to the above technical solution, in step 2, step 3 and step 4, if the AUV underwater robot is low on power, the AUV underwater robot returns to the ROV underwater robot for charging.

According to the above technical solution, the disease database also includes a model obtained by preprocessing of public data sets, modification of configuration files, and training. The data sets include corrosion, weathering, cracks, exposed steel bars, and spalling. Video frames containing diseases are imported into the model for processing to obtain a second detection result.

According to the above technical solution, the disease database includes n sub-directories, and the n sub-directories divide the water diversion tunnel into "Section 1", "Section 2" ... "Section n" on average according to the length of the water diversion tunnel, and the directory name is the divided area; in step 2, step 3 and step 4, the detection results of the AUV underwater robot in each section are stored in the directory of the corresponding section, and the data in each section is output to a new window after being processed by the model, that is, the detection results of the first, second and abnormal data are presented.

According to the above technical solution, in step two, the AUV underwater robot uses a multi-beam three-dimensional scanning sonar to conduct a 360° mobile and continuous scanning survey of the water diversion tunnel, and generates high-density three-dimensional point cloud data of the tunnel in real time.

According to the above technical solution, in step three, three-dimensional laser scanning technology is used to analyze the tunnel characteristics and combine the "precise denoising technology of tunnel wall radial density" to divide the tunnel into micro-elements along the axis and the circumference of the tunnel according to the geometric structure. The micro-element fitting residuals are calculated to analyze the relative deformation of the tunnel wall, so as to achieve millimeter-level ranging accuracy and obtain high-density point cloud data of the entire tunnel wall within the effective measuring range.

According to the above technical solution, in step three, when the near-optical detection component approaches the disease point for optical photography, the heading data is output by the laser gyroscope installed in the AUV underwater robot, and the heading data is converted into an angle with the center of the orbit as the reference center. The angle of each point on the orbit is converted into a one-to-one corresponding plane coordinate; the depth of the camera is obtained based on the depth sensor in the cabin, and the disease is accurately located by combining the depth and the plane coordinates.

According to the above technical solution, in step 4, the location where the abnormal situation occurs is photographed, and the detection personnel determine whether there is a disease and the type of disease.

According to the above technical solution, five identifiable diseases are stored in each section, and the video frames of the identified diseases are saved in the disease folder corresponding to the section subdirectory; when all sections of a tunnel are identified, the number of disease images stored in each section is summarized. When the storage number of a section exceeds the set value, it means that the area is seriously damaged and needs to be repaired; the identification number of each of the five types of diseases in a specific section is compared with the set threshold to further determine the type of disease that needs to be repaired immediately.

The present invention has the following beneficial effects:

1. A rapid survey is conducted with acoustic inspection as the main method and optical inspection as the auxiliary method. At the same time as the rapid survey with optical inspection as the auxiliary method, a detailed inspection of the defective points in the survey process is carried out through close optical photography and laser ranging. The present invention adopts a "survey + detailed inspection" detection method, which integrates a variety of acoustic and optical sensors and verifies each other to ensure that the detection accuracy is not lost. It solves the technical problems of traditional underwater detection methods such as single means, low efficiency and large errors. The defect recognition accuracy reaches the centimeter level, the detection information is comprehensive and reliable, and the safety risk is low, which not only improves the detection speed but also ensures the detection accuracy.

2. During the detection process, the ROV underwater robot is connected to the external shore-based power cable as a transfer station. During the detection process, it provides power replenishment and positioning for the AUV underwater robot at any time, which improves the endurance of the AUV underwater robot and the accuracy of detection.

3. Divide the underwater tunnel into sections, store five identifiable diseases in each section, and save the video frames of the identified diseases in the disease folder corresponding to the section subdirectory; when all sections of a tunnel are identified, summarize the number of disease image storages in each section. When the number of storages in a section exceeds the set value, it means that the area is seriously damaged and needs to be repaired; compare the identification number of each of the five types of diseases in a specific section with the set threshold to further determine the type of disease that needs to be repaired immediately; provide general solutions based on the type of disease, including main materials, technical preparations, and construction plans, which simplifies the processing of detection data and makes it easy to display the detection results.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of the structure of an embodiment of the present invention;

FIG2 is an appearance diagram of an AUV underwater robot according to an embodiment of the present invention;

FIG3 is an outline diagram of an ROV underwater robot according to an embodiment of the present invention;

FIG4 is a schematic structural diagram of a first power module according to an embodiment of the present invention;

FIG5 is a schematic diagram of the structure of a near optical detection assembly according to an embodiment of the present invention;

FIG6 is a schematic diagram of the structure of a mechanical claw according to an embodiment of the present invention;

7 is a schematic diagram of the structure of an ROV underwater robot according to an embodiment of the present invention;

FIG8 is a schematic structural diagram of a second power module according to an embodiment of the present invention;

FIG9 is a detection flow chart of an embodiment of the present invention;

10 is a flowchart of constructing a disease database according to an embodiment of the present invention;

In the figure, 1. AUV underwater robot; 2. ROV underwater robot; 3. Cable; 4. Shore-based power supply; 5. External controller; 6. Acoustic 3D hologram; 7. Laser scanner; 8. Near-optical detection component; 8-1. Telescopic arm; 8-2. Searchlight; 8-3. Camera; 8-4. Lifting structure; 9. First hull; 10. Second hull; 11. Water pump; 12. Balance wing; 13. Steering rudder; 14. Auxiliary propeller; 15. Two propellers, one large and one small; 16. Mechanical arm; 17. Mechanical claw; 19. Cabin; 20. Hatch cover; 21. Fixed frame; 22. Docking rod; 23. Docking port; 24. Chassis frame; 25. Driving wheel; 26. Driven wheel; 27. Rubber track; 28. Tunnel.

DETAILED DESCRIPTION

The present invention is described in detail below with reference to the accompanying drawings and embodiments.

As shown in Figures 1 to 10, the present invention adopts an intelligent detection mechanism for long-distance water diversion tunnel 28 during operation, including an AUV underwater robot 1 for detection and an ROV underwater robot 2 for providing energy supply and information exchange, and the two work in cooperation with each other; the AUV underwater robot is connected to a shore-based power supply 4 and an external controller 5 by a cable 3, the AUV underwater robot is carried on the ROV underwater robot, and the external controller controls the deployment and recovery of the AUV underwater robot through the ROV underwater robot, and the two are connected by wireless signals. A detection module is provided on the inner side of the head of the AUV underwater robot, and the detection module includes an acoustic 3D hologram 6 for sonar detection, a laser scanner 7 for laser scanning, and a near-optical detection component 8 for optical detailed detection; the ROV underwater robot is provided with a sonar signal receiving device for receiving the sonar signal of the AUV underwater robot and realizing the positioning function according to the received sonar signal.

Further, the AUV underwater robot includes a first hull 9, on which a first power module, a disease identification module, and a first docking and fixing module are also provided; the power module is provided at the periphery of the outer wall of the first hull, the first docking and fixing module and the docking module are provided at the front of the first hull, the disease identification module is provided inside the first hull, and the virus identification module is electrically connected to the detection module;

The ROV underwater robot includes a second hull 10, on which are arranged a second power module for driving the ROV underwater robot to move, a second docking and fixing module matching the first docking and fixing module, an inductive charging module for providing energy to the ROV underwater robot, and a cable module; the second power module is arranged at the bottom of the second hull; the first docking and fixing module is connected to the second docking and fixing module, and the ROV underwater robot and the AUV underwater robot are combined and separated under the action of the first and second docking and fixing modules; the inductive charging module is arranged on the inner side of the second docking and fixing module, and is electrically connected to the AUV underwater robot when the first and second docking and fixing modules are docked; one end of the cable module is connected to the ROV underwater robot, and the other end is connected to the shore-based power supply.

Furthermore, the acoustic 3D hologram uses underwater high-resolution tunnel structure detection multi-beam profile sonar. In this embodiment, it is equipped with the T2250-360 Tunnel Profiler measurement system developed by Laure of the United States; it uses high-frequency, low-power multi-beam technology to obtain continuous 360° profile data and continuous high-resolution underwater tunnel wall images, with clear imaging, forming a real-time three-dimensional point cloud model of the tunnel structure, and can perform visual damage assessment and measurement on three planes. It has fast data acquisition speed and high spatial resolution, extracts typical sections of the tunnel to view structural deformation, and detects sediment conditions at the same time. The scanning section error reaches 0.029m, and the detection effect is good, which provides a guarantee for the acquisition of fine data of the final tunnel. It can also be used for navigation, collision avoidance, regional search, real-time tracking and other applications.

The laser scanner uses an underwater three-dimensional laser scanner. This embodiment is equipped with the ULS-500 underwater three-dimensional laser scanner developed by Canada's 2G Robotics. It uses three-dimensional laser scanning technology, namely real-scene replication technology, and uses a three-dimensional laser scanner as a carrier. It uses the principle of laser ranging and adopts a non-contact high-speed laser measurement method. Through the analysis of tunnel characteristics and combined with the "precise denoising technology of radial density of tunnel wall", the whole is divided into micro-elements along the axis and the circumference of the tunnel according to the geometric structure, and the micro-element fitting residual is calculated to analyze the relative deformation of the tunnel wall, so as to achieve millimeter-level ranging accuracy and obtain high-density point cloud data of the entire tunnel wall within the effective measurement range. After denoising, splicing, coloring and other processing, a deformation analysis color chromatogram is formed, and any section of the tunnel can be extracted and queried separately, and cracks in the tunnel wall and the shape of the section at any position can be automatically extracted. A three-dimensional digital information model of the inner wall defect information of the tunnel can be established to realize three-dimensional digital management. The sound wave information uploaded and returned by the AUV underwater robot can also be processed by the ROV underwater robot to realize the real-time positioning of the AUV underwater robot.

The near optical detection assembly includes a telescopic arm 8-1, a searchlight 8-2 and a camera 8-3. One end of the telescopic arm is fixed on the AUV underwater robot, and the other end is connected to the searchlight and the camera. The camera is located inside the searchlight. The telescopic arm has three degrees of freedom, which can realize lighting and searchlighting at multiple angles. The camera adopts an underwater camera, and the searchlight adopts a high-brightness LED light source, and the brightness of the light source is smoothly adjusted. In this embodiment, the underwater camera adopts two UM-30 series cameras of UNIQ (UM-301 series is a near-infrared (NIR) camera) and the camera is equipped with two high-brightness LED light sources. The shutter speed of the UM-301 series camera is 1/60～1/31000s, the minimum illumination is 0.03lux, and it has an asynchronous acquisition function, which allows the camera to acquire high-quality images without any blur at a very high shutter speed. The camera's follow-up LED light source can provide 4000Lm of illumination, and the light source brightness adjustment is smooth. The 100-level brightness adjustment can achieve quasi-stepless brightness adjustment; it is small in size and light in weight, and can be easily mounted on the AUV underwater robot without affecting the posture and operation of the AUV underwater robot. The excellent hardware of the camera lighting system ensures the excellent performance of the robot in underwater imaging.

Furthermore, the connection between the telescopic arm and the AUV underwater robot is set as a lifting structure 8-4, under the action of the lifting structure, the near optical detection component is stored in the AUV underwater robot body. When not in use or under specific circumstances, the searchlight, camera and mechanical arm can be stored in the body to protect it and make it easy to carry. After storage, the volume is reduced and the resistance is reduced.

Furthermore, the first power module includes a forward part, a steering part and a snorkeling part. The forward part is arranged at the tail of the first hull, the steering part is arranged at the rear side of the bottom end of the first hull, the steering part is arranged at the front end of the forward part, and the snorkeling part is arranged on both sides of the first hull.

Furthermore, the forward part includes a water pump 11 and a balance wing 12. A cavity running through the front and back is provided on the first hull, the water pump is provided in the cavity, and the balance wings are evenly spaced on the outer wall of the cavity. The water pump is used to pump seawater around the first hull into the hull, and then the seawater is sprayed behind the first hull to obtain forward power, which has good power redundancy to cope with complex and changeable underwater turbulent environments. In this embodiment, evenly distributed balance wings are designed on the outside of the water pump, and the balance wings play a role in stabilizing the hull and controlling the stability of the forward direction.

The steering part includes a steering rudder 13 and an auxiliary propeller 14. The auxiliary propeller is arranged directly behind the steering rudder. The steering rudder plays the role of steering the AUV underwater robot forward, and the auxiliary propeller plays the role of assisting steering and providing power for the AUV underwater robot to move backward.

The snorkeling part is symmetrically arranged on both sides of the first hull, and two propellers 15, one large and one small, are arranged on one side. The larger propeller is arranged on the side of the first hull near the head, and the smaller propeller is arranged on the side of the first hull near the tail. The propeller is used to control the buoyancy and diving of the AUV underwater robot; the design of using two propellers, one large and one small, on one side at the same time allows the two propellers to achieve different output powers through different rotation speeds and output powers, which can ensure the buoyancy and diving rates while also improving the positioning accuracy of the AUV underwater robot, so that it can operate smoothly at a certain depth. At the same time, the design of multiple propellers also greatly improves the reliability of the operation of the AUV underwater robot. When a propeller loses power, the other three propellers can still ensure the safe operation of the AUV underwater robot and achieve the buoyancy and diving operations.

Further, the first fixed docking module includes a mechanical arm 16, a mechanical claw 17, and a cabin 19 symmetrically arranged on both sides of the front of the first cabin; the cabin is arranged inside the first hull, and a hatch 20 is provided at the contact point between the cabin and the outer wall of the first hull, and one end of the mechanical arm is arranged in the cabin; the other end of the mechanical arm is connected to the mechanical claw, and the mechanical arm and the mechanical claw have two states of being retracted into the cabin and extending out of the cabin. The mechanical claw has four mechanical fingers, which cooperate with each other to clamp the fixed rod to play a fixing role. In this embodiment, the mechanical claw has five degrees of freedom in five directions and can realize rotation and extension in multiple directions. The mechanical claw has four mechanical fingers, which cooperate with each other to clamp the fixed rod to play a fixing role. During the normal navigation of the AUV underwater robot, the mechanical arm is folded and retracted into the hull, and the hatch is isolated from the outside to play a protective role; at the same time, the design that can be retracted into the boat can reduce the resistance of the AUV underwater robot when it is navigating underwater, which is convenient for transportation. When the AUV underwater robot needs to return to the ROV underwater robot to recharge and dock before charging, the two robotic arms extend and unfold. There is a searchlight on both sides of the robotic arms for lighting and assisting the mechanical claws at the head to fix the first hull on the ROV underwater robot.

The second fixed docking module includes a fixing frame 21 arranged on both sides of the front of the second cabin and an electromagnetic lock arranged inside the second cabin. The fixing frame is a funnel mesh structure, and the smaller end of the fixing frame is fixed at the front of the second cabin; the ROV underwater robot and the AUV underwater robot are combined and separated by grasping and releasing the fixing frame by a mechanical claw; the electromagnetic lock realizes the locking function according to the status of the ROV underwater robot and the AUV underwater robot.

During the docking process, the ROV underwater robot controls the photoconductive switch to guide the optical navigation of the AUV underwater robot, and at the same time, the AUV underwater robot turns on the camera to observe the docking situation. When the AUV underwater robot successfully enters the ROV underwater robot, the docking rod containing the charging coil is inserted into the power transmission docking port of the ROV underwater robot. The AUV underwater robot is fixed by the mechanical arm and the mechanical claw using the internal fixing frame of the ROV underwater robot. Then the AUV underwater robot mechanical arm and the funnel mesh fixing frames on both sides of the ROV underwater robot are effectively linked, triggering the electromagnetic lock, thereby locking the AUV underwater robot in the ROV underwater robot. After confirming that the AUV underwater robot is locked in the docking station, data exchange and power transmission can be carried out. When the AUV underwater robot is detached, the electromagnetic lock is controlled to unlock, and then the AUV underwater robot starts a new task.

The heading data is output by the laser gyroscope installed in the AUV underwater robot, and converted into an angle with the center of the orbit as the reference center. The angle of each point on the orbit is converted into a one-to-one plane coordinate. The depth of the camera is obtained based on the depth sensor in the cabin, and the disease is accurately located by combining the depth and plane coordinates.

Furthermore, a docking rod 22 containing a charging coil is provided at the front of the AUV underwater robot, and a docking port 23 matching the docking rod is provided on the charging induction module of the ROV underwater robot; the docking rod is provided between the two mechanical arms, and the docking port is provided in the middle of the fixed frame, and the charging rod and the docking port are combined and separated by grasping and releasing the fixed frame by the mechanical claw. The docking of the charging rod and the power transmission docking port realizes non-contact power transmission, which is an important system that greatly improves the endurance of the AUV underwater robot.

Furthermore, the second power module adopts a crawler structure for underwater operations, which can provide better traction performance on various surfaces; it includes a chassis frame 24, which is arranged at the bottom of the second cabin, and is provided with a DC motor, a driving wheel 25, a driven wheel 26, and a rubber crawler 27 on the chassis frame, all of which are made of corrosion-resistant materials such as stainless steel, and the rubber crawler adopts an elastic belt with tread. A weight reduction scheme was adopted in the design to control the weight of the entire crawler crawling chassis within 60 kilograms. Compared with conventional crawler crawling chassis on the market, it has the characteristics of light weight and compact structure. Taking into account the need to overcome obstacles, the crawler part is installed directly below the ROV underwater robot to ensure that the crawler robot can crawl stably at the bottom of the tunnel and reach the designated location smoothly.

Furthermore, the cable module uses an umbilical cable, which is reinforced with Kevlar material and has neutral buoyancy. The all-copper cable structure has good durability and can simultaneously perform functions such as communication and power transmission. After the AUV underwater robot is docked with the ROV underwater robot, the shore-based power can be used through the cable to charge and recharge the AUV underwater robot during work.

Based on the above device, the present invention provides an intelligent detection method for long-distance water diversion tunnel operation period diseases, including an AUV underwater robot for detection and an ROV underwater robot for providing energy replenishment and information exchange. The method includes the following steps:

Step 1: Under the control of the operating console, the ROV underwater robot loads the AUV underwater robot and enters the entrance of the water diversion tunnel. At the entrance, the ROV underwater robot deploys the AUV underwater robot;

Step 2: The AUV underwater robot performs an acoustic 3D scan of the entire tunnel to obtain a complete 3D image of the diversion tunnel structure and determine the defect points, and save the first detection results;

Step 3: Perform laser scanning on the entire tunnel again to obtain complete high-density point cloud data of the entire tunnel wall and determine the defect points. Use the near-optical detection component to take close-up optical images of the defect points scanned twice, match and compare the video frames containing the defects with the defect database, and save and analyze the second detection results.

Step 4: If there are still abnormal conditions that cannot be determined whether there is a disease after two scans, the AUV underwater robot will take a video of the location of the abnormal condition and store it in the disease database. If there is no abnormal condition that cannot be determined whether there is a disease, it will directly proceed to step 5;

Step 5: The AUV underwater robot returns to the ROV underwater robot, and the ROV underwater robot carries the AUV underwater machine and is recovered by the console.

The defect recognition module mainly reads the defect points scanned by the acoustic 3D hologram and laser scanner, then takes a video of the defect points by approaching the optical detection component, and then imports the video frames containing the defect points into the defect database for matching and comparison to obtain the detection results. The maintenance staff can determine the type of defect based on the filmed content and arrange subsequent maintenance optimization work through data collation and analysis.

A rapid survey is adopted with acoustic inspection as the main method and optical inspection as the auxiliary method. At the same time as the rapid survey with optical inspection as the auxiliary method, a detailed inspection of the defective points in the survey process is carried out through close-range optical photography and laser ranging. The present invention adopts a "survey + detailed inspection" detection method, which integrates a variety of acoustic and optical sensors and verifies each other to ensure that the detection accuracy is not lost, and solves the technical problems of traditional underwater detection means such as single means, low efficiency, and large errors. The defect recognition accuracy reaches the centimeter level, the detection information is comprehensive and reliable, and the safety risk is low, which not only improves the detection speed but also ensures the detection accuracy.

Furthermore, in step 2, step 3 and step 4, if the AUV underwater robot is low on power, the AUV underwater robot returns to the ROV underwater robot for charging.

Furthermore, the disease database also includes a model obtained by preprocessing the public data set, modifying the configuration file, and training. The data set includes corrosion, weathering, cracks, exposed steel bars, and spalling; the video frames containing the disease are imported into the model for processing to obtain the second detection result. This device intends to use the yolov5 model. The yolov5 code is open source in the github software. Download the source code and modify the code in the pycharm editor according to the detection requirements.

The source of the dataset is public, from the CVPR19 paper "Meta-learning Convolutional Neural Architectures for Multi-target Concrete Defect Classification with the COncrete DEfect BRidge IMage Dataset", which contains six categories: Corrosion Stain, Efflorescence, Crack, Exposed Bars, Spallation. Since the images in the dataset are not completely uniform, the dataset is manually preprocessed so that the image files correspond to the label files one by one, and the label files (.xml) are converted to text files (.txt) for the model to use. Since the images in the dataset are not completely uniform, the dataset is manually preprocessed so that the image files correspond to the label files one by one, and the label files (.xml) are converted to text files (.txt) for the model to use. After the parameter configuration is completed, the model starts training. After the training is completed, the trained weight file and model are loaded to test the recognition effect of the disease images. During identification, the algorithm calculates the predicted probabilities of the five major diseases, selects the disease type corresponding to the maximum predicted probability, and screens the detection frame with higher confidence by comparing the predicted probability with the set threshold. Modifying the data source can call video data from different cameras. In order to test the degree of video jamming, the data source is changed to the mobile phone camera IP address. During the test, the video window output is relatively smooth, without obvious jamming.

Furthermore, the disease database includes n sub-directories, and the n sub-directories divide the water diversion tunnel into "Section 1", "Section 2" ... "Section n" on average according to the length of the water diversion tunnel, and the directory name is the divided area; in step 2, step 3 and step 4, the detection results of the AUV underwater robot in each section are stored in the directory of the corresponding section, and the data in each section is output to a new window after being processed by the model, that is, the detection results of the first, second and abnormal data are presented.

Furthermore, in step two, the AUV underwater robot uses a multi-beam three-dimensional scanning sonar to conduct a 360° mobile and continuous scanning survey of the water diversion tunnel, and generates high-density three-dimensional point cloud data of the tunnel in real time.

Furthermore, in step 2, three-dimensional laser scanning technology is used to analyze the characteristics of the tunnel and combine it with the "precise denoising technology of radial density of the tunnel wall" to divide the tunnel into micro-elements along the axis and the circumference of the tunnel according to the geometric structure, and the micro-element fitting residual is calculated to analyze the relative deformation of the tunnel wall, so as to achieve millimeter-level ranging accuracy and obtain high-density point cloud data of the entire tunnel wall within the effective measurement range. When the camera recognizes a defect, the video frame where the defect is located will be stored in the defect database of the specified path to evaluate the degree of the defect of the detected water diversion tunnel, so as to provide corresponding maintenance and optimization measures.

Furthermore, in step three, when the near-optical detection component approaches the diseased point for optical photography, the heading data is output by the laser gyroscope installed in the AUV underwater robot, and is converted into an angle with the center of the orbit as the reference center. The angle of each point on the orbit is converted into a one-to-one corresponding plane coordinate; the depth of the camera is obtained based on the depth sensor in the cabin, and the disease is accurately located by combining the depth and the plane coordinates.

Furthermore, in step 4, the location of the abnormal situation is photographed, and the detection personnel determine whether there is a disease and the type of disease.

Furthermore, five identifiable diseases are stored in each section, and the video frames of the identified diseases are saved in the disease folder corresponding to the subdirectory of the section; when all sections of a tunnel are identified, the number of disease images stored in each section is summarized. When the number of storages in a section exceeds the set value, it means that the area is seriously damaged and needs to be repaired; the number of identifications of the five types of diseases in a specific section is compared with the set threshold to further determine the type of disease that needs to be repaired immediately. According to the type of disease, a general solution is provided, including main materials, technical preparation, and construction plan.

Additional functions of this device:

1. Hybrid robot’s underwater acoustic positioning technology

The long baseline system is composed of a transponder array, a data signal processor, an acoustic transceiver, and a transponder; the transponder array is composed of more than three transponders placed on the seabed. The relative formation of the transponder array of the system must be repeatedly measured and confirmed to ensure that the formation error is minimized, thereby improving positioning accuracy.

2 Navigation technology of hybrid robot

The intelligent underwater robot navigation system provides the vehicle with information such as position, heading, depth, speed and attitude to ensure the safety of navigation and operation of the underwater robot. When multiple underwater robots work together, they should also have collaborative positioning and collaborative navigation capabilities. The sensor-based navigation method is that when the intelligent underwater robot is navigating underwater, it relies on the sensors inside the system to navigate without receiving external signals. This navigation method can be divided into strapdown inertial navigation system, dead reckoning, terrain matching, gravity field navigation, etc. This embodiment adopts terrain navigation. The environment inside the long water diversion tunnel is complex, and there is water flow all year round. It is impossible to navigate in a conventional way. The robot detects the terrain and landforms in the tunnel, thereby generating a route and traveling along it. This ensures that it can choose different routes by observing the terrain and landforms in different environments, so as to meet the requirements of adapting to various environments.

3. The obstacle avoidance and auxiliary positioning functions of the hybrid robot. The ROV underwater robot adopts underwater multi-source combined navigation technology, and realizes underwater high-precision navigation based on inertial navigation, omnidirectional sonar, depth meter, image, etc. The underwater navigation system includes the alignment, combined navigation, installation error correction, fault detection and other technologies of the underwater robot navigation system. The navigation system uses high-precision inertial measurement units, omnidirectional navigation sonars, high-precision depth meters and other measurement sensors to form a dedicated underwater robot positioning system, providing underwater positioning data support for automatic driving, defect detection, and post-maintenance. The software design realizes the comprehensive operation of multiple data sources obtained by asynchronous, different physical quantities, and different measurement principles to obtain the optimal positioning data.

The above are only preferred embodiments of the present invention, which certainly cannot be used to limit the scope of rights of the present invention. Therefore, equivalent changes made according to the scope of the patent application of the present invention still fall within the protection scope of the present invention.

Claims (7)

1. An intelligent detection method for diseases in the operation period of a long-distance diversion tunnel comprises an AUV underwater robot for detection and an ROV underwater robot for energy supply and information interaction; the method is characterized in that: the method comprises the following steps:

Step one: the ROV underwater robot is controlled by an operation table to load the AUV underwater robot into the entrance of the diversion tunnel, and the AUV underwater robot is distributed at the entrance;

step two: carrying out acoustic 3D scanning on the whole tunnel by an AUV underwater robot to obtain a complete 3D image of the diversion tunnel structure, judging disease points and storing a first detection result;

Step three: performing laser scanning on the whole tunnel again to obtain complete full-tunnel wall high-density point cloud data, judging disease points, performing optical imaging on the disease points scanned twice by a low-beam optical detection assembly, matching and matching a video frame containing diseases with a disease database, and storing and analyzing a second detection result;

in the third step, a three-dimensional laser scanning technology is adopted, the integral tunnel is divided into infinitesimal parts along the axis and the circumferential direction of the tunnel according to the geometric structure by analyzing the characteristics of the tunnel and combining with an accurate denoising technology of the radial density of the tunnel wall, and the infinitesimal fitting residual error is calculated to analyze the relative deformation of the tunnel wall, so that the millimeter-level ranging precision is realized, and the high-density point cloud data of the full tunnel wall in the effective range is obtained;

outputting course data through a laser gyroscope arranged in the AUV underwater robot when the near-light detection assembly shoots the disease point near light, converting the course data into angles by taking the circle center of the track as a reference center, and converting the angles of each point on the track into one-to-one corresponding plane coordinates; according to a depth sensor in the cabin, the depth of the camera is obtained, and accurate positioning of diseases is realized through combination of the depth and plane coordinates;

Step four: if the abnormal condition which cannot be judged whether the disease exists after the two scans, the AUV underwater robot shoots the position where the abnormal condition exists and stores the position into a disease database, and if the abnormal condition which cannot be judged whether the disease exists, the step five is directly carried out;

step five: the AUV underwater robot returns to the ROV underwater robot, and the ROV underwater robot carries the AUV underwater robot and is operated and recovered by a control console.

2. The intelligent detection method for the diseases in the long-distance diversion tunnel operation period according to claim 1 is characterized by comprising the following steps: in the second, third and fourth steps, if the electric quantity of the AUV underwater robot is insufficient, the AUV underwater robot returns to the ROV underwater robot for charging.

3. The intelligent detection method for the diseases in the long-distance diversion tunnel operation period according to claim 1 is characterized by comprising the following steps: the disease database also comprises a model obtained by preprocessing, modifying configuration files and training of a public data set, wherein the data set comprises corrosion, weathering, cracks, exposed steel bars and peeling; and importing the video frames containing the diseases into a model for processing, thereby obtaining a second detection result.

4. The intelligent detection method for the diseases in the operation period of the long-distance diversion tunnel according to claim 3, which is characterized in that: the disease database comprises n subdirectories, wherein the n subdirectories divide tunnels into a section 1, a section 2, … … and a section n according to the lengths of diversion tunnels, and the directory names are divided areas; in the second, third and fourth steps, the detection results of the AUV underwater robot in each section are stored in the catalog of the corresponding section, and the data in each section are output to a newly built window after being processed by a model, namely the detection results of the first, second and abnormal data are displayed.

5. The intelligent detection method for the diseases in the long-distance diversion tunnel operation period according to claim 4 is characterized in that: in the second step, the AUV underwater robot adopts multi-beam three-dimensional scanning sonar to conduct 360-degree mobile continuous scanning and census on the diversion tunnel, and high-density three-dimensional point cloud data of the tunnel are generated in real time.

6. The intelligent detection method for the diseases in the long-distance diversion tunnel operation period according to claim 4 is characterized in that: and step four, shooting the position where the abnormal condition is located, and judging whether the disease exists or not and the type of the disease according to the detection personnel.

7. The intelligent detection method for the diseases in the long-distance diversion tunnel operation period according to claim 5, which is characterized in that: under each section, five identifiable diseases are respectively stored, and video frames of the identified diseases are stored in a disease folder corresponding to the subdirectory of the section; after all sections of a certain tunnel are identified, collecting the storage number of disease images under each section, and when the storage number of a certain section exceeds a set value, indicating that the disease of the area is serious and maintenance is needed; and comparing the identification number of each of the five diseases in the specific section with a set threshold value, and further judging the type of the disease to be immediately maintained.