WO2022156192A1 - 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法 - Google Patents

用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法 Download PDF

Info

Publication number
WO2022156192A1
WO2022156192A1 PCT/CN2021/111595 CN2021111595W WO2022156192A1 WO 2022156192 A1 WO2022156192 A1 WO 2022156192A1 CN 2021111595 W CN2021111595 W CN 2021111595W WO 2022156192 A1 WO2022156192 A1 WO 2022156192A1
Authority
WO
WIPO (PCT)
Prior art keywords
detection
wall
climbing robot
rapid
geological radar
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2021/111595
Other languages
English (en)
French (fr)
Inventor
王正方
王静
康文强
刘涵池
万玉壮
隋青美
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shandong University
Original Assignee
Shandong University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN202110071015.7A external-priority patent/CN112917483B/zh
Priority claimed from CN202110706142.XA external-priority patent/CN113447536B/zh
Application filed by Shandong University filed Critical Shandong University
Priority to US17/765,215 priority Critical patent/US12072298B2/en
Publication of WO2022156192A1 publication Critical patent/WO2022156192A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J11/00—Manipulators not otherwise provided for
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B62—LAND VEHICLES FOR TRAVELLING OTHERWISE THAN ON RAILS
    • B62D—MOTOR VEHICLES; TRAILERS
    • B62D57/00—Vehicles characterised by having other propulsion or other ground- engaging means than wheels or endless track, alone or in addition to wheels or endless track
    • B62D57/02—Vehicles characterised by having other propulsion or other ground- engaging means than wheels or endless track, alone or in addition to wheels or endless track with ground-engaging propulsion means, e.g. walking members
    • B62D57/024—Vehicles characterised by having other propulsion or other ground- engaging means than wheels or endless track, alone or in addition to wheels or endless track with ground-engaging propulsion means, e.g. walking members specially adapted for moving on inclined or vertical surfaces
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N15/00—Investigating characteristics of particles; Investigating permeability, pore-volume or surface-area of porous materials
    • G01N15/08—Investigating permeability, pore-volume, or surface area of porous materials
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/04—Analysing solids
    • G01N29/043—Analysing solids in the interior, e.g. by shear waves
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/22—Details, e.g. general constructional or apparatus details
    • G01N29/225—Supports, positioning or alignment in moving situation
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/22—Details, e.g. general constructional or apparatus details
    • G01N29/26—Arrangements for orientation or scanning by relative movement of the head and the sensor
    • G01N29/265—Arrangements for orientation or scanning by relative movement of the head and the sensor by moving the sensor relative to a stationary material
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N29/00—Investigating or analysing materials by the use of ultrasonic, sonic or infrasonic waves; Visualisation of the interior of objects by transmitting ultrasonic or sonic waves through the object
    • G01N29/44—Processing the detected response signal, e.g. electronic circuits specially adapted therefor
    • G01N29/4454—Signal recognition, e.g. specific values or portions, signal events, signatures
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/881—Radar or analogous systems specially adapted for specific applications for robotics
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/885—Radar or analogous systems specially adapted for specific applications for ground probing
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S13/00—Systems using the reflection or reradiation of radio waves, e.g. radar systems; Analogous systems using reflection or reradiation of waves whose nature or wavelength is irrelevant or unspecified
    • G01S13/88—Radar or analogous systems specially adapted for specific applications
    • G01S13/89—Radar or analogous systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01S—RADIO DIRECTION-FINDING; RADIO NAVIGATION; DETERMINING DISTANCE OR VELOCITY BY USE OF RADIO WAVES; LOCATING OR PRESENCE-DETECTING BY USE OF THE REFLECTION OR RERADIATION OF RADIO WAVES; ANALOGOUS ARRANGEMENTS USING OTHER WAVES
    • G01S17/00—Systems using the reflection or reradiation of electromagnetic waves other than radio waves, e.g. lidar systems
    • G01S17/88—Lidar systems specially adapted for specific applications
    • G01S17/89—Lidar systems specially adapted for specific applications for mapping or imaging
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/023—Solids
    • G01N2291/0232—Glass, ceramics, concrete or stone
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/02—Indexing codes associated with the analysed material
    • G01N2291/028—Material parameters
    • G01N2291/0289—Internal structure, e.g. defects, grain size, texture
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N2291/00—Indexing codes associated with group G01N29/00
    • G01N2291/26—Scanned objects
    • G01N2291/269—Various geometry objects
    • G01N2291/2698—Other discrete objects, e.g. bricks

Definitions

  • the invention belongs to the field of non-destructive testing, and in particular relates to a wall-climbing robot system and method for fast non-destructive testing of hidden defects of culverts and gates.
  • a sluice gate is a low-head water retaining and discharge project built on river channels and embankments. In flood season, it cooperates with river embankments and drainage and water storage projects to control water flow.
  • culverts there will be construction defects of concrete, such as voids, cracks and other appearance quality defects.
  • concrete carbonization, steel corrosion, cracks, chloride ion erosion and other diseases often occur.
  • Most hydraulic structures, including culverts and sluices are aging and have increasingly prominent diseases during use. There are serious safety hazards in culverts and sluices, and the number of dangerous culverts and sluices is huge.
  • the existing detection methods for the safety of culvert gates include rebound method, ultrasonic-rebound method and core drilling method, all of which need to be done manually, but there are drawbacks: on the one hand, manual detection is time-consuming and laborious, and there are certain dangers. , on the other hand, it will cause damage to the culvert gate.
  • the culvert gate can be detected by installing sensors, but it can only detect the local part of the culvert gate, and cannot realize all-round detection.
  • the existing ground penetrating radar, ultrasonic detector and other equipment have been applied to the internal structure of concrete detection, and can achieve good results. The method of using robots to carry instruments and equipment for detection is in constant development.
  • Identifying hidden defects in culverts and gates based on geological radar is the key link of the system.
  • technologies such as signal processing, image processing, and artificial intelligence are used at home and abroad for automatic interpretation based on geological radar profiles.
  • the research mainly focuses on the target recognition based on ground radar images.
  • the recognition methods include Hough transform, wavelet transform, support vector machine, clustering, convolutional neural network and other methods, which can locate the category and approximate position of the target, but cannot estimate the target. the exact shape of the body.
  • the hyperbolic waveform reflected by the same target is not only related to the geological radar data of the corresponding spatial position of the target, but also related to the surrounding geological radar data.
  • the reflected waveform of a target especially the target located at the cropping boundary, is likely to be truncated, resulting in the reflection of the target in the local short line B-Scan.
  • the waveform is incomplete or there is interference. In this case, processing and splicing B-Scan results of short survey lines will lead to discontinuity of target recognition results of continuous survey lines, and it is difficult to process GPR data of continuous survey lines of any length.
  • the existing method separates the inversion of the permittivity and the identification of the disease, and cannot realize the inversion of the permittivity and the identification of the disease type at the same time.
  • both the inversion of the dielectric constant of concrete and the accurate identification of the disease can be completed, and there is a certain correlation between the two.
  • the existing methods do not fully consider the relationship between the two tasks, and only realize the inversion or identification of a single task, and cannot fully exploit the mutual promotion relationship, and realize the two tasks of inversion and identification at the same time.
  • Shandong University proposed in its patent document "An Intelligent Inversion Method of Ground Penetrating Radar Based on Deep Learning" (Patent Application No.: 202010723091.7, Application Date: 2020.01.08, Application Publication No. CN111781576A)
  • the deep learning-based GPR intelligent inversion method can reconstruct the subsurface dielectric constant distribution end-to-end using GPR data.
  • this method is suitable for processing fixed-length short-line GPR data, and in the process of continuous long-line radar data inversion, it is easy to cause the numerical discontinuity and shape dislocation of the target inversion of the continuous line.
  • this method is also used to realize the single function of inverting the dielectric constant based on the ground penetrating radar data.
  • the detection technology for hidden defects of culverts and sluices mainly used at present mainly has the following problems: (1) At present, the detection of hidden defects of culverts and sluices is mainly manual detection, and automatic detection cannot be realized. It is difficult to accurately judge whether the culvert is safe or not, which is time-consuming and labor-intensive. (2) At present, the ultrasonic probe used for ultrasonic detection of concrete can only realize fixed position detection, which is difficult to realize mobile detection, and the detection efficiency is low. (3) At present, the detection of steel bar corrosion inside culvert gates mainly adopts the half-cell potential method, which has high environmental requirements and low detection efficiency.
  • the present invention provides a wall-climbing robot system and method for rapid non-destructive detection of hidden defects of culverts and gates. detection.
  • a first aspect of the present invention provides a wall-climbing robot system for rapid non-destructive detection of hidden defects in culverts.
  • a wall-climbing robot system for rapid non-destructive detection of hidden defects of culverts and gates comprising:
  • the navigation and positioning system and the mobile system are installed on the robot body;
  • the mobile ultrasonic rapid detection system and the rust detection system are installed at the bottom of the vehicle body;
  • the dual power system consists of a non-contact negative pressure adsorption device and a rotor booster device.
  • the non-contact negative pressure adsorption device is installed at the bottom of the car body, and the rotor booster device is installed on both sides of the car body;
  • the master controller communicates with the navigation and positioning system, the mobile system, the automatic knock detection system, the mobile ultrasonic rapid detection system, the rust detection system and the dual power system respectively.
  • a second aspect of the present invention provides a working method of a wall-climbing robot system for rapid non-destructive detection of hidden defects in culverts.
  • a working method of a wall-climbing robot system for rapid non-destructive detection of hidden defects of culverts and gates comprising:
  • the navigation and positioning system completes the path planning, enters the fast census mode, starts the automatic knock detection system and the rust detection system, and the mobile system moves at a high speed, using the automatic knock detection system, the rust detection system, the multi-eye panoramic camera and the navigation and positioning system, Realize rapid census of defect areas and rough identification and location of defects;
  • the mobile system of the wall-climbing robot moves to the defect position at high speed. After reaching the defect position, the mobile ultrasonic rapid inspection is selectively enabled according to the type of defect. system and corrosion detection system, and the mobile system is changed to low-speed movement.
  • the general controller controls the rotation and expansion device according to the pressure sensor information, which is used to maintain the same and stable bonding force of each ball-type ultrasonic probe.
  • the wall-climbing robot moves to another defect position at high speed, and moves at a low speed during the detection process. After all defects are detected, the wall-climbing robot returns to the origin to complete the fine inspection of the defect area;
  • the data obtained based on ultrasound is processed by the recognition method based on deep learning, the deep neural network based on deformable convolution rotation area detection is used to identify the location and category of internal defects, and the "inversion-recognition" multi-task deep neural network that fuses time series information is used. Internal lesions are continuously imaged from multiple sets of ultrasound detection data.
  • the working method of the wall-climbing robot system for fast and non-destructive detection of hidden defects of culverts and gates also includes inverting the dielectric constant of concrete and identifying diseases at the same time.
  • the specific process is as follows:
  • the real radar detection data is used for permittivity inversion and disease identification.
  • the mobile ultrasonic rapid detection system of the present invention realizes mobile ultrasonic testing by adopting a ball-type ultrasonic probe.
  • the ball-type ultrasonic probe can be attached to the measured surface, and monitored by a pressure sensor.
  • Real-time bonding force by controlling the rotating telescopic device, the bonding force of each ultrasonic probe is kept stable, and the consistency of the measurement data is realized; and in the ball-type ultrasonic probe, the universal moving ball is combined with the ultrasonic probe to realize the movement.
  • Ultrasonic detection the universal moving ball is coupled with the ultrasonic probe through the soft coupling block, the transmission of ultrasonic waves is realized, and the coupling force is increased by the elastic device, which reduces the loss of ultrasonic energy.
  • the pre-automatic knock detection system of the present invention realizes automatic knock detection by using an electromagnetic knock device, and controls the coordination of two electromagnets through the automatic knock controller to realize knocks of different frequencies and different strengths.
  • the sound receiver receives the sound signal, and realizes the mobile automatic tapping by cooperating with the moving wheel.
  • the dual power system of the present invention provides the main bonding force of the robot through the non-contact negative pressure adsorption device, provides auxiliary power through the rotor booster system, and provides the bonding force information fed back by the pressure sensor and the torque sensor through the master controller. Control, when the air pressure provided by the non-contact negative pressure adsorption device reaches the upper limit, but still does not meet the requirements of the fit force, the rotor booster device is turned on to achieve power supplementation.
  • the working method of the wall-climbing robot system for the rapid non-destructive detection of hidden defects of culvert gates of the present invention realizes all-round detection of culvert gates by dividing the detection process into rapid general testing and fine testing, and is determined by rapid general testing
  • the specific position of the hidden defect according to the defect position obtained by the rough detection, realizes the fine detection of the hidden defect, which can not only realize the all-round detection of the culvert, but also ensure the efficiency and quality of the detection.
  • the present invention innovatively provides a concrete dielectric constant inversion and disease identification method for continuous survey line geological radar data of arbitrary length, which uses the deep neural network model to realize the detection data of the continuous survey line of arbitrary length by the geological radar.
  • Automated inversion and identification This method uses a combination of convolutional neural network and recurrent neural network for context feature extraction and interaction, and makes full use of the correlation between B-Scan data of local short survey lines at different positions of continuous survey lines to improve the consistency of disease characteristics at the splicing point, and solves the problem of splicing.
  • the discontinuity problem of disease shape and dielectric constant value at the splicing position is suitable for the data processing of continuous survey lines of arbitrary lengths.
  • the present invention fully exploits the interdependence between the inversion and identification tasks of the geological radar, and simultaneously realizes the inversion of the dielectric properties and the accurate identification of the disease type, location and shape by using a network structure, and realizes the inversion of the geological radar.
  • the feature sharing between the performance task and the recognition task improves the generalization ability of the model.
  • the present invention obtains the data pair of "Ground Radar Profile - Dielectric Constant Distribution Map & Target Category Label" by means of simulation, and can obtain sufficient dielectric constant distribution by combining various background media and disease-filled media
  • the training data of the image and target category labels; the Rake wavelet with the same frequency and phase as the actual GPR wavelet is used as the source wavelet for the forward modeling of the simulation data, and the simulation data is preprocessed to make the simulation ground radar detection data closer to reality
  • the detection data of geological radar provides a guarantee for the generalization ability of subsequent models on real detection data of geological radar.
  • the present invention also obtains real geological radar detection data of continuous survey lines of any length, and uses data enhancement technology to construct a training data set of real geological radar data. And adopt the transfer learning method, use the real training data set to fine-tune the intelligent inversion and recognition training network model based on simulation data training, so that the intelligent inversion and recognition network model can learn the distribution of real geological radar detection data, and can more accurately identify the real Concrete structure disease.
  • the method proposed in the present invention can be used in the fields of concrete non-destructive testing, road disease detection, engineering geological survey, etc., and realizes the fine inversion of the internal dielectric properties of the structure and the type, location and shape of the disease based on the detection of the continuous survey line of any length. accurate identification.
  • FIG. 1 is an overall block diagram of a wall-climbing robot system for fast non-destructive detection of hidden defects of culverts and gates according to an embodiment of the present invention
  • FIG. 2 is a sectional view of a wall-climbing robot system for rapid non-destructive detection of hidden defects of culverts and gates according to an embodiment of the present invention
  • FIG. 3 is a bottom view of a wall-climbing robot system for rapid non-destructive detection of hidden defects of culverts and gates according to an embodiment of the present invention
  • FIG. 4 is a schematic structural diagram of an automatic tapping device according to an embodiment of the present invention.
  • FIG. 5 is a schematic structural diagram of a mobile ultrasonic probe according to an embodiment of the present invention.
  • FIG. 6 is a flowchart of a method for inversion of the dielectric constant of concrete and disease identification for continuous survey line geological radar data of arbitrary length according to Embodiment 1;
  • Embodiment 7 is a schematic diagram of the intelligent inversion and identification network structure of the continuous survey line geological radar of arbitrary length shown in Embodiment 1;
  • FIG. 9 is a schematic diagram of an interactive structure of a long survey line spatiotemporal feature according to an embodiment
  • Fig. 10 is the simulation ground-based radar detection data shown according to the embodiment.
  • FIG. 11 is a simulated dielectric constant distribution diagram according to an embodiment
  • Fig. 13 is the dielectric constant distribution diagram of the intelligent inversion and identification prediction of the continuous survey line geological radar of arbitrary length shown in the embodiment;
  • FIG. 14 is a target class label predicted by the intelligent inversion and identification network of the continuous survey line geological radar of arbitrary length according to the embodiment.
  • the wall-climbing robot system for rapid non-destructive detection of hidden defects of culverts and gates of this embodiment includes a robot body; a navigation and positioning system and a mobile system are installed on the robot body; automatic tapping The detection system is installed at the front end of the vehicle body; the mobile ultrasonic rapid detection system and the rust detection system are both installed at the bottom of the vehicle body; the dual power system consists of a non-contact negative pressure adsorption device and a rotor booster device.
  • the negative pressure adsorption device is installed at the bottom of the car body, and the rotor booster device is installed on both sides of the car body; the general controller, which is connected with the navigation and positioning system, mobile system, automatic knock detection system, mobile ultrasonic rapid detection system, and corrosion detection system and dual power systems communicate with each other respectively.
  • the mobile ultrasonic rapid detection system adopts a ball-type ultrasonic probe array to realize the floating coupling and mobile detection of ultrasonic contacts.
  • the balls are made of metal, ceramics and other materials with good sound permeability.
  • the inner wall of the spherical chamber is equipped with balls 1407, which can improve the rolling smoothness of the balls.
  • a soft coupling block 1402 is used to connect the balls and the ultrasonic probe.
  • One end of the coupling block is bonded to the acoustic emission end (or receiving end) of the ultrasonic probe 1403, and the other end has a spherical groove for fitting with the universal ball.
  • the soft coupling block can be made of materials with small sound attenuation coefficients such as polygrease soft plastic, neoprene rubber, cast polyurethane rubber, etc., which can be selected by those skilled in the art according to the actual situation.
  • several ball-type ultrasonic probes 14 are installed on the mobile ultrasonic probe mounting frame 15 on the rear side of the robot to form an ultrasonic array.
  • the upper part of the ball-type ultrasonic probe is connected to a pressure sensor 16 for monitoring real-time bonding force, and the other end of the pressure sensor is connected to
  • the third elastic device 17 is used to provide the bonding force, and the other end of the elastic device is equipped with a rotary telescopic device 23 to adjust the bonding force.
  • the resultant force is the same and stable.
  • the ultrasonic detection controller controls the transmission and reception of ultrasonic signals of each ultrasonic probe, and the received information is input to the general controller for processing.
  • the automatic knock detection system adopts an electromagnetic knocking device 3, which includes a knocking hammer 301 for knocking, and the other end of the knocking hammer is connected to the first electromagnet 304 and installed on the casing of the knocking device Inside, the second electromagnet 305 is fixed on the upper part of the shell 302 of the knocking device, and the two electromagnets are controlled by the automatic knocking controller to cooperate with the first elastic device 306 and the second elastic device 307 inside to realize automatic knocking, electromagnetic
  • the knocking device and the ranging encoder 4 are installed on the middle moving wheel in the front of the robot, the ranging encoder 4 is used for positioning, and the sound receiver 5 is installed on the rear side of the automatic knocking device close to the ground.
  • the automatic knocking detects The system completes the rapid detection of defective areas.
  • the striking device housing 302 is provided with a mounting plate 303 .
  • the corrosion detection system adopts a two-step staggered detection mode based on geological radar to observe the corrosion state of steel bars in concrete.
  • the geological radar 13 is installed on the rear side of the bottom of the robot body.
  • the first step is rough measurement. Walk in the direction, obtain the radar survey line profile (B-Scan), and use the deep neural network-based steel bar fast positioning algorithm to determine the steel bar arrangement direction; the second part performs corrosion diagnosis, scans along the steel bar layout direction, and uses different positions.
  • the characteristic parameters such as the amplitude of the time domain signal (A-Scan) change, and the rust degree and rust position of the current position are judged.
  • the dual power system includes a non-contact negative pressure adsorption device 19 and a rotor booster device 10.
  • the non-contact negative pressure adsorption device is installed in the middle part of the robot body and is equipped with an air pressure sensor 20, so
  • the rotor booster device is installed on both sides of the robot body, and the torque sensor 9 is installed on one end of the rigid-flexible arc rod 8 of the mobile system connected to the car body to monitor the real-time bonding force. Both the negative pressure adsorption device and the rotor booster device are used.
  • the master controller controls whether to start the rotor booster according to the information of the torque sensor 9 and the air pressure sensor 20.
  • the rotor booster 10 includes a rotor motor 11 and a rotor 12 , and the rotor motor 11 is used to drive the rotor 12 to rotate.
  • the non-contact negative pressure adsorption device is powered by the negative pressure adsorption motor 18 .
  • the moving system includes a rigid-flexible arc rod, a wheel, a drive motor and a torque sensor.
  • a rigid-flexible arc rod 8 is used to connect the robot body 1 and the wheels for shock absorption and transmission of the bonding force.
  • the drive motor 7 is connected to the moving wheel 2 to realize the movement and steering of the robot, and a torque sensor 9 is installed at one end of the connection body. , which is used to monitor the real-time fit force of the robot.
  • the middle moving wheel connecting rod 6 is connected between the two moving wheels 2 at the front of the robot.
  • the navigation and positioning system includes a multi-eye panoramic camera 21 and a lidar 22 , the multi-eye panoramic camera is used to detect surface diseases of the culvert gate, and the lidar is used to realize the automatic composition of the culvert gate detection area.
  • the working principle of the wall-climbing robot system for rapid non-destructive detection of hidden defects of culverts and gates of the present embodiment is to adopt the detection mode of "quick census ⁇ fine detection", and the specific steps include:
  • Step (1) start the robot dual power system to make the robot adsorb on the side of the quilt.
  • the dual power system monitors the robot's bonding force in real time during the detection process.
  • the bonding force provided by the non-contact negative pressure adsorption device 19 meets the requirements, It is not necessary to open the rotor booster device 10, on the contrary, it is necessary to open the rotor booster device to realize the supplementary fit force;
  • Step (2) The navigation and positioning system completes the path planning, enters the quick census mode, activates the automatic knock detection system and the rust detection system, the mobile system moves at a high speed, and the electromagnetic knocking device 3 knocks the side of the victim at a fixed frequency, and the sound
  • the receiver 5 receives the sound signal, while the radar transmits and receives electromagnetic waves, and the miniature multi-eye panoramic camera 21 shoots the surface of the detection area to realize a quick census of the defect area.
  • the received information is processed by the general controller to complete the rough identification of defects and position;
  • Step (3) According to the defect positioning information obtained in the quick census mode, re-plan the path and enter the fine detection mode.
  • the mobile system of the wall-climbing robot moves to the defect position at high speed. After reaching the defect position, it is selectively turned on according to the type of defect.
  • the mobile ultrasonic rapid detection system and the rust detection system, and the mobile system is changed to low-speed movement.
  • the general controller controls the rotation and expansion device 23 according to the information of the pressure sensor 16, which is used to keep each ball-type ultrasonic probe.
  • the adhesion force is the same and stable. Every time a defect detection is completed, the wall-climbing robot moves to another defect position at high speed, and moves at a low speed during the detection process. detection;
  • the data obtained based on ultrasound is processed by the recognition method based on deep learning, the deep neural network based on deformable convolution rotation area detection is used to identify the location and category of internal defects, and the "inversion-recognition" multi-task deep neural network that fuses time series information is used. Internal lesions are continuously imaged from multiple sets of ultrasound detection data.
  • the defect shapes of different concrete structures are complex and diverse. However, due to the influence of diffraction, reflection, scattering, etc., the defect response in the ultrasonic image and the actual shape of the concrete defect are often not one-to-one correspondence.
  • this embodiment constructs an intelligent identification method for concrete defect response and position based on deformable convolution, and introduces a deformable convolution that can arbitrarily change the shape according to the defect shape. Instead of the traditional regular shape convolution, the shape of the event axis of different defects is adaptively extracted, so as to find and locate the defects distributed in any direction.
  • a multi-task deep neural network of "inversion-recognition" fused with time series information is designed to solve the continuous imaging problem of internal concrete defects under long measurement lines.
  • this method designs a spatiotemporal feature extraction module to fuse the spatiotemporal information of measurement data at different times.
  • the test data at multiple times are input into the coding-decoding network in parallel, and secondly, the decoded features are input in parallel into the ConvLSTM-based spatiotemporal feature extraction module, the correlation between the ultrasound data before and after the time is mined, and the measurement results at the previous time are memorized and merged.
  • the multi-task branch output is designed, and the defect type and wave speed distribution are output at the same time, which realizes the integration of defect identification and wave speed inversion.
  • the working method of the wall-climbing robot system for fast and non-destructive detection of hidden defects of culverts and gates further includes inverting the dielectric constant of concrete and identifying diseases at the same time, as shown in FIG. 6 , and the specific process is as follows:
  • Step S1 establish a simulation training data set.
  • the step S1 specifically includes:
  • Step S101 constructing the dielectric constant distribution maps and target category labels of tunnel lining structures with various lengths.
  • a dielectric constant distribution map of the cross section of lining structure is generated according to each combination method.
  • a pair of target type labels corresponding to the profile permittivity profile is generated according to each target body type it contains.
  • the tunnel lining structure model includes lining models with lengths of 5m, 10m, 15m, etc., to simulate a variety of continuous survey line lengths in line with reality.
  • the background medium includes various background media such as plain concrete and reinforced concrete, and the disease includes cavity, non-compact, crack, void, fault, karst cave, etc., and the internal medium of the disease is water, air, mud, rock and other media.
  • the target body category is further divided into: steel bar, background, no water cavity, no water compact, no water crack, no water void, no water fault, no water karst cavity, no water cavity, In order to better identify the target body type in the tunnel lining
  • Step S102 Using the rake wavelet whose frequency and phase are consistent with the actual GPR wavelet as the source wavelet for the simulation data modeling, forward modeling is performed on each permittivity distribution map to generate a corresponding GPR profile, and It is preprocessed by methods such as direct wave removal, gain adjustment, and noise addition to improve the adaptability of real data.
  • the forward modeling adopts the FDTD method.
  • the direct wave removal is to use the obtained simulated geological radar data to subtract the data of the path where the waveform without disease is located; the gain adjustment is collected by comparing the rake wavelet and the geological radar equipment used to generate the simulated geological radar data against the air.
  • the amplitude level of the air-mining direct wave is adjusted to the gain of the simulation data so that it is in the same dimension; the noise is added by collecting the geological radar data of the background medium at different sites, and randomizing from 0 to 1 Select different scales to add to the simulated GPR data.
  • Step S103 Obtain multiple sets of “Ground Radar Profiles” through the obtained corresponding processed GPR profiles (as shown in FIG. 10 ), the corresponding dielectric constant distribution maps (as shown in FIG. 11 ) and target category labels (as shown in FIG. 12 ).
  • - Dielectric constant distribution map & target class label" data pair, the dielectric constant distribution map and target class label data in each data pair are used as the label of the geological radar profile, and a simulation training data set is established.
  • Step S2 Build and train the intelligent inversion and identification network model of the continuous survey line geological radar of any length
  • the step S2 specifically includes:
  • Step S201 Build an intelligent inversion and identification network structure for continuous survey line geological radar of arbitrary length
  • the intelligent inversion and identification network structure of the continuous survey line of any length of the geological radar adopts a cascade of "spatial feature extraction structure of local short survey line ⁇ spatial and temporal feature interaction structure of long survey line ⁇ dual task structure of dielectric constant inversion and target recognition" way of implementation.
  • the intelligent inversion and identification network structure of the arbitrary-length continuous survey line geological radar is based on the arbitrary length continuous survey line geological radar detection data D n (n ⁇ [1,N] (N is all the continuous survey lines in the data set.
  • a sliding window with a step size of S 1 is used to cut the D n sequence into a local short survey line B-Scan profile sequence (K is the number of B-Scan sections of fixed-size local short measuring lines after D n is cut).
  • the local short-line spatial feature extraction structure performs parallel B-Scan profiles for each local short-line Perform spatial context feature extraction to obtain feature sequence
  • the long-line spatiotemporal feature interaction structure is As input, through bidirectional spatiotemporal information fusion within the feature sequence, extract the enhanced feature sequence containing relatively stable target features (including shape, class, and permittivity)
  • the dual-task structure of permittivity inversion and target recognition processes each local short line B-Scan profile in parallel Enhanced features of Reconstruct the corresponding permittivity distribution map and identifying target class labels
  • the overlapping positions are averaged to obtain the permittivity map P n and the target class label In for the entire continuous line .
  • the spatial feature extraction structure of local short survey lines is realized by multiple parallel fully convolutional network structures.
  • Each fully convolutional network structure independently processes each local short line B-Scan profile extracted from different locations in a continuous line of arbitrary length
  • the spatial context feature extraction is performed on it to form the feature sequence of the detection data of the continuous survey line of any length.
  • the fully convolutional network structure is implemented by using the DenseUnet network.
  • the DenseUnet structure consists of one encoding path and one decoding path.
  • the encoding path includes 4 convolutional layers, 4 max pooling layers and 4 dense block structures
  • the decoding path includes 4 volumes Convolutional layers, 4 transposed convolutional layers and 4 dense block structures, with a convolutional layer connection between the encoder and decoder paths.
  • the encoding path compresses the geological radar detection data into high-level abstract features through stepwise downsampling, and the decoding path obtains a preliminary representation of the interior spatial structure of the tunnel lining through stepwise upsampling operations.
  • the size of the convolution kernel of the convolution layer structure is 3*3, the step size is 1, and the number of channels is 64, so as to extract the spatial features in the geological radar detection data.
  • the maximum pooling layer is connected after each dense block of the encoding path.
  • the maximum pooling layer has a kernel size of 2*2 and a stride of 2.
  • the downsampled output is sent to the next dense block of the encoding path.
  • the transposed convolutional filter size is 4*4
  • the stride is 2
  • the number of channels is 64.
  • the upsampled output is channel-wise concatenated with the dense block from the corresponding encoding path, and then fed into the next dense block of the decoding path.
  • the dense block structure includes 3 convolutional layer structures connected in series.
  • the convolutional layer convolution kernel size is 3*3, the stride is 1, and the number of channels is 64.
  • Each convolutional layer uses the output of all previous convolutional layers. Information is added to its own output to facilitate the flow of gradients and better learn the representation of GPR data.
  • a skip layer connection is adopted between the corresponding dense block structures in the encoding path and the decoding path, which can transfer the spatial information in the encoding process to the decoding process of the inner structure of the tunnel lining.
  • the local short survey line spatial feature extraction structure performs spatial context feature extraction on the B-Scan profile data of each local short survey line at different positions extracted from the continuous survey line of any length, and forms the geological radar profile data to the spatial information of the concrete internal structure. Preliminary mapping.
  • the spatial-temporal feature interaction structure of the long survey line is implemented by a one-layer Bi-ConvLSTM structure.
  • the Bi-ConvLSTM realizes the local short line B-Scan profile feature sequence Bidirectional spatiotemporal information fusion in Enhanced Feature Sequence for Spatial Context Information Between Line B-Scan Profiles for inversion or identification of tunnel lining interior targets with continuous shape, class, and permittivity values.
  • the long-line spatiotemporal feature interaction structure includes a layer of Bi-ConvLSTM structure.
  • the Bi-ConvLSTM contains a forward ConvLSTM layer, a reverse ConvLSTM layer, and a convolutional layer.
  • the feature sequence of geological radar data extracted by the Bi-ConvLSTM with the local short-line spatial feature extraction structure As input, the forward ConvLSTM layer passes the hidden state forward along the GPR detection direction between each local short line B-Scan profile and memory cells The backward ConvLSTM layer transfers the hidden state backwards between each local short line B-Scan profile along the reverse of the GPR detection direction and memory cells For each imported local short line profile feature
  • the Bi-ConvLSTM unit converts the corresponding forward hidden state and the backward hidden state Concatenated in the channel dimension, and then passed through a convolutional layer to obtain the final enhanced features containing the spatial context information of the local short-line B-Scan profile.
  • the ConvLSTM inner convolution structure and convolution layer convolution structure both use 3*3 convolution kernels, the stride is 1, and the number of channels is 64.
  • the spatiotemporal feature interaction structure of the long survey line fuses the bidirectional spatiotemporal information between the B-Scan data features of the local short survey lines at different positions extracted from the continuous survey lines of any length, to ensure the inversion results of the dielectric constant of the tunnel lining structure of the continuous survey lines. and the continuity and accuracy of target recognition results.
  • each parallel two-branch convolutional network structure processes each local short line B-Scan profile independently Enhanced Feature Extraction via Spatial-Temporal Feature Interaction Structure of Long Survey Line
  • the dielectric constant distribution of the tunnel lining structure is reconstructed and the type, location and contour of the disease are identified.
  • the two-branch convolutional network structure includes an inversion task branch and a recognition task branch, and each branch includes a convolution layer and an activation function.
  • the inversion task branch first adopts 1 convolution layer to reduce the input feature dimension, the convolution kernel size is 1*1, the stride is 1, and the number of channels is 1.
  • the recognition task branch first adopts 1 convolution layer to reduce the input feature dimension, the convolution kernel size is 1*1, the stride is 1, and the number of channels is 9.
  • a sigmoid activation function is connected to the tail of the inversion branch to regress the permittivity distribution map, and the recognition branch is connected to a softmax function to classify the target class.
  • the dual-task structure of permittivity inversion and target identification simultaneously reconstructs the permittivity distribution of concrete structures and identifies disease types, locations and contours, so as to realize feature sharing between the inversion task and the identification task of the geological radar, and enhance the generalization ability of the model.
  • Step S202 Based on the simulation training data set, train the intelligent inversion and identification network model of the continuous survey line geological radar of any length
  • the loss function combined with Mean Square Error (MAE), Structural Similarity Index (SSIM) and Lovasz_Softmax is used, and ADAM optimization algorithm is used to optimize the intelligent inversion and identification network model of continuous survey line geological radar of arbitrary length. , training to obtain the intelligent inversion and recognition network model of the continuous survey line geological radar of arbitrary length.
  • MAE Mean Square Error
  • SSIM Structural Similarity Index
  • Lovasz_Softmax Lovasz_Softmax
  • Step S3 establishing a real training data set, and adopting the transfer learning method to obtain an intelligent inversion and identification network model of the geological radar with any length of continuous survey line suitable for the actual detection data of the geological radar.
  • the step S3 specifically includes:
  • Step S301 Establish a real training data set
  • Step S302 Based on the real training data set, a transfer learning method is used to obtain an intelligent inversion and identification network model of the geological radar with any length of continuous survey line suitable for the actual detection data of the geological radar.
  • Step S4 Use the intelligent inversion and identification network model of the continuous survey line geological radar of any length to perform dielectric constant inversion and disease identification on the actually collected continuous survey line geological radar detection data, and obtain the corresponding permittivity distribution map and target. type label.
  • the background medium of the lining structure, the disease form, the filling medium in the disease, and the disease category measured by the continuous survey line can be restored, so as to achieve the purpose of disease detection.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Chemical & Material Sciences (AREA)
  • Biochemistry (AREA)
  • Pathology (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Immunology (AREA)
  • Analytical Chemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Electromagnetism (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Mechanical Engineering (AREA)
  • Acoustics & Sound (AREA)
  • Robotics (AREA)
  • Transportation (AREA)
  • Combustion & Propulsion (AREA)
  • Dispersion Chemistry (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Automation & Control Theory (AREA)
  • Investigating Or Analyzing Materials By The Use Of Ultrasonic Waves (AREA)
  • Manipulator (AREA)

Abstract

用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法。该机器人系统包括机器人车体（1）；导航定位系统和移动系统，均安装于机器人车体（1）上；自动敲击检测系统，其安装于车体（1）前端；移动式超声快速检测系统和锈蚀检测系统；双动力系统，由非接触式负压吸附装置和旋翼助推装置组成，非接触式负压吸附装置安装于车体（1）底部，旋翼助推装置安装于车体（1）两侧；总控制器，其与导航定位系统、移动系统、自动敲击检测系统、移动式超声快速检测系统、锈蚀检测系统和双动力系统分别相互通信。该系统能够通过超声数据识别内部缺陷位置与类别，并对任意长度连续测线地质雷达数据进行混凝土介电常数反演和病害识别。

Description

用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法 技术领域
本发明属于无损检测领域，尤其涉及一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法。
背景技术
本部分的陈述仅仅是提供了与本公开相关的背景技术信息，不必然构成在先技术。
水闸是修建在河道、堤防上的一种低水头挡水、泄水工程。汛期与河道堤防和排水蓄水工程配合，发挥控制水流的作用。在涵闸建设时，会出现混凝土的施工缺陷，例如空洞，裂缝等外观质量缺陷，在涵闸使用过程中往往会出现混凝土的碳化、钢筋锈蚀、裂缝、氯离子侵蚀等病害，随着时间的推移，包括涵闸在内的多数水工建筑物在使用过程中老化、病害情况日益突出，涵闸安全隐患严重，病险涵闸数量庞大，为预防灾难的发生，亟需对涵闸内隐蔽缺陷进行全方位检测。
已有的对涵闸安全的检测方式有回弹法、超声-回弹综合法及钻芯法等，且均需要人工完成，但其存在弊端：一方面人工检测既费时又费力并且存在一定的危险，另一方面会对涵闸造成损伤。除此之外可通过安装传感器对涵闸进行检测，但只能对涵闸的局部进行检测，并不能实现全方位的检测，目前已有的探地雷达，超声探测仪等设备已应用于混凝土内部结构的探测，并且可到达很好的效果。采用机器人搭载仪器设备进行检测的方式处于不断发展之中，例如：河海大学在其申请的专利文献“一种基于巨磁电阻元件的闸门检测机器人及检测方法”(专利申请号：201811173150.7，申请日：2018.10.09，申请公布号CN109324112A)提出了将磁探伤传感器安装于检测机器人之上，用于检测闸门表面的缺陷，但其并不能检测闸门内部空洞、钢筋锈蚀等病害，且巡检速度较慢；河海大学在其申请的专利文献“一种基于相控阵超声波探伤仪的闸门检测装置及检测方法”(专利申请号：201811172523，申请日：2018.10.09，申请公布号CN109324121A)中设计的检测机器人搭载有超声波相控阵矩阵探头，并可进行扫射式检测，但其不能够完成快速普测，检测效率低；山东大学在其申请的专利文献“用于桥隧结构病害无损检测诊断的自动爬墙式雷达光电机器人系统”(专利申请号：201810565136，申请日：2018.06.04，申请公布号CN108731736A)中利用旋翼系统的反推力使机器人紧贴于桥隧结构表面进行内部病害的检测，该技术由于反推力有限，难以搭载多种设备。
基于地质雷达识别涵闸内隐蔽缺陷是系统的关键环节。伴随着新一代信息技术的快速发 展，国内外将信号处理、图像处理、人工智能等技术用于基于地质雷达剖面图的自动解释。研究主要围绕基于地质雷达图像的目标识别展开，识别方法包括霍夫变换、小波变换、支持向量机、聚类、卷积神经网络等方法，可以定位目标体的类别和大致位置，但不能估计目标体的准确形态。并且，采用上述方法进行识别均存在以下问题：(1)地质雷达通常采用“边走边测”工作模式形成一条长测线地质雷达数据，而在地质雷达数据中，地下结构内部的目标体的反射呈现近似双曲线特性，同一目标体反射的双曲线波形不仅仅与目标体所对应空间位置的地质雷达数据有关，而且和其周围的地质雷达数据相关。当对数据进行裁剪后分成短测线B-Scan后，某一目标体特别是位于裁剪边界位置的目标体，其反射波形极可能被截断，导致在局部短测线B-Scan中目标体反射波形不完整或存在干扰。在此情况下对短测线B-Scan结果进行处理进而拼接，会导致连续测线的目标体识别结果的不连续，难以处理任意长度连续测线的地质雷达数据。(2)现有方法将介电常数的反演与病害识别过程割裂，不能同时实现介电常数反演与病害类型识别。而实际上，基于同一组地质雷达数据既可以实现混凝土介电常数的反演，也可以完成对病害的准确识别，并且二者存在一定相关性。但现有方法没有充分考虑两个任务之间的关联关系，仅实现反演或者识别单一任务，无法充分挖掘其相互促进关系，同时实现反演与识别两个任务。例如：山东大学在在其申请的专利文献“一种基于深度学习的探地雷达智能反演方法”(专利申请号：202010723091.7,申请日：2020.01.08,申请公布号CN111781576A)中提出的一种基于深度学习的探地雷达智能反演方法，可以利用地质雷达数据端到端地重建地下介电常数分布。但是该方法适用于处理固定长度的短测线地质雷达数据，在连续长测线雷达数据反演过程中容易造成连续测线的目标体反演的数值不连续和形状错位。而且该方法也用于实现了基于探地雷达数据反演介电常数这一单一功能，没有在实现介电常数反演的同时对目标体的类型进行识别。
发明人发现，目前主要采用的涵闸隐蔽缺陷检测技术主要存在以下问题：(1)目前对涵闸隐蔽缺陷的检测以人工检测为主，不能实现自动检测，存在检测效率低并且难以对涵闸进行大面积细致性检测，难以准确地判断涵闸是否安全，既耗时又费力。(2)目前用于混凝土超声探测的超声探头只能实现固定位置检测，难以实现移动式检测，检测效率低。(3)目前对涵闸内部钢筋锈蚀的检测，主要采用半电池电位法，对环境要求高，检测效率低，只能完成抽样检测，难以完成对涵闸内钢筋锈蚀的全方位自动检测。(4)采用上述方法进行反演与识别会导致连续测线的目标体成像结果的不连续或反演的数值不连续，难以处理任意长度连续测线的地质雷达数据，而且现有方法将介电常数的反演与病害识别过程割裂，不能同时实现介电常数反演与病害类型识别。
发明内容
为了解决上述背景技术中存在的技术问题，本发明提供一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法，其利用机器人搭载雷达和超声探头，能够对涵闸隐蔽缺陷进行全方位的检测。
本发明的第一个方面提供一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统。
一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，包括：
机器人车体；
导航定位系统和移动系统，均安装于机器人车体上；
自动敲击检测系统，其安装于车体前端；
移动式超声快速检测系统和锈蚀检测系统，均安装于车体底部；
双动力系统，由非接触式负压吸附装置和旋翼助推装置组成，非接触式负压吸附装置安装于车体底部，旋翼助推装置安装于车体两侧；
总控制器，其与导航定位系统、移动系统、自动敲击检测系统、移动式超声快速检测系统、锈蚀检测系统和双动力系统分别相互通信。
本发明的第二个方面提供一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法。
一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，包括：
启动双动力系统，使机器人吸附于被侧面，双动力系统在检测过程中实时监测机器人的贴合力；
导航定位系统完成路径规划，进入快速普查模式，启动自动敲击检测系统和锈蚀检测系统，移动系统以高速度移动，利用自动敲击检测系统、锈蚀检测系统、多目全景相机和导航定位系统，实现对缺陷区域的快速普查以及缺陷的粗略识别和定位；
根据快速普查模式所获得的缺陷定位信息，重新进行路径规划，进入精细检测模式，爬壁机器人移动系统以高速移动至缺陷位置，到达缺陷位置后，根据缺陷的类型选择性开启移动式超声快速检测系统和锈蚀检测系统，同时移动系统改为低速移动，移动式超声快速检测系统运行时，总控制器根据压力传感器信息控制旋转伸缩装置，用于保持各滚珠式超声探头的贴合力相同且稳定，每完成一处缺陷检测爬壁机器人均以高速移动至另一缺陷位置，检测过程中低速移动，待所有缺陷检测完毕，爬壁机器人返回原点，完成对缺陷区域的精细检测；
基于超声获取的数据采用基于深度学习的识别方法进行处理，采用基于可变形卷积旋转区域检测深度神经网络识别内部缺陷位置与类别，采用融合时序信息的“反演-识别”多任务深度神经网络从多组超声探测数据中对内部病害连续成像。
进一步地，所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，还包括同时反演混凝土介电常数及识别病害，其具体过程为：
对不同应用场景的地质雷达设备提取实际发射子波信号，将其作为仿真建模源子波信号进行正演，生成相应的地质雷达剖面图，其同建立的介电常数分布图和目标类别标签形成数据对，构成仿真训练数据集；
构建地质雷达智能反演与识别模型，基于仿真训练数据集，训练所述地质雷达智能反演与识别模型；
建立真实训练数据集，基于真实训练数据集和迁移学习方法微调所述地质雷达智能反演与识别模型；
利用微调后的地质雷达智能反演与识别模型对真实雷达检测数据进行介电常数反演和病害识别。
与现有技术相比，本发明的有益效果为：
(1)本发明的移动式超声快速检测系统通过采用滚珠式超声探头，实现移动式超声检测，通过安装架中的弹性装置，实现了滚珠式超声探头与被测面贴合，通过压力传感器监测实时贴合力，通过控制旋转伸缩装置，保持了各超声探头贴合力稳定，实现了测量数据的一致性；而且在滚珠式超声探头中，通过将万向移动滚珠与超声探头相结合，实现了移动式超声探测，通过软质耦合块将万向移动滚珠与超声探头耦合，实现了超声波的传递，通过弹性装置增加耦合力，减少了超声波能量的损失。
(2)本发明的前置自动敲击检测系统，通过采用电磁敲击装置，实现自动敲击检测，通过自动敲击控制器控制两电磁铁配合，实现了不同频率不同力度的敲击，通过声音接收器接收声音信号，通过与移动轮配合，实现了移动式自动敲击。
(3)本发明的双动力系统，通过非接触式负压吸附装置，提供机器人主要贴合力，通过旋翼助推系统，提供辅助动力，通过总控制器根据压力传感器和力矩传感器反馈的贴合力信息进行控制，当非接触负压吸附装置所提供的气压达到上限，但仍不满足贴合力需求时，开启旋翼助推装置，实现动力补充。
(4)本发明的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，通过将检测过程分为快速普测和精细检测，实现了对涵闸的全方位检测，通过快速普测确定隐蔽缺陷的具体位置，根据粗略检测所获得的缺陷位置，实现了对隐蔽缺陷的精细检测，既能实现对涵闸全方位检测，又能保证检测的效率和质量。
(5)本发明创新性提供了一种用于任意长度连续测线地质雷达数据的混凝土介电常数反演和病害识别方法，利用深度神经网络模型对任意长度连续测线的地质雷达检测数据实现 自动化反演和识别。该方法采用卷积神经网络与递归神经网络相结合进行上下文特征提取与交互，充分利用连续测线不同位置局部短测线B-Scan数据之间关联关系提升拼接处病害特征的一致性，解决了拼接位置病害形态与介电常数数值的不连续问题，适用于任意长度连续测线地质雷达数据处理。
(6)本发明充分挖掘地质雷达反演与识别任务的相互依赖关系，利用一个网络结构同时实现了同时实现了介电特性反演与病害类型、位置及形状的准确识别，实现了地质雷达反演任务和识别任务的特征共享，提升了模型泛化能力。
(7)本发明通过模拟仿真方式获取“地质雷达剖面图-介电常数分布图&目标类别标签”数据对，通过采用多种背景介质和病害填充介质进行组合，能够得到充分的介电常数分布图和目标类别标签训练数据；采用与实际地质雷达子波频率和相位一致的雷克子波作为仿真数据正演的源子波，并对仿真数据进行预处理，使得仿真地质雷达检测数据更贴近真实地质雷达检测数据，为后续模型在真实地质雷达检测数据上的泛化能力提供了保障。
(8)本发明还获取了任意长度连续测线的真实地质雷达检测数据，利用数据增强技术构建了真实地质雷达数据训练数据集。并采用了迁移学习方法，利用真实训练数据集微调基于仿真数据训练的智能反演与识别训练网络模型，使得智能反演与识别网络模型学习到真实地质雷达检测数据分布，能够更准确的识别真实混凝土结构病害。
(9)本发明提出的方法能够用于混凝土无损检测、道路病害检测、工程地质勘察等领域，实现基于任意长度连续测线探测的结构内部介电特性的精细反演与病害类型、位置及形状的准确识别。
本发明附加方面的优点将在下面的描述中部分给出，部分将从下面的描述中变得明显，或通过本发明的实践了解到。
附图说明
构成本发明的一部分的说明书附图用来提供对本发明的进一步理解，本发明的示意性实施例及其说明用于解释本发明，并不构成对本发明的不当限定。
图1为本发明实施例的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的整体框图；
图2为本发明实施例的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统剖面图；
图3为本发明实施例的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统仰视图；
图4为本发明实施例的自动敲击装置结构示意图；
图5为本发明实施例的移动式超声探头结构示意图；
图6为根据实施例一所示的用于任意长度连续测线地质雷达数据的混凝土介电常数反演与病害识别方法的流程图；
图7为根据实施例一所示的任意长度连续测线地质雷达智能反演与识别网络结构示意图；
图8为根据实施例一所示的基于DenseUnet的局部短测线空间特征提取结构示意图；
图9为根据实施例所示的长测线时空特征交互结构示意图
图10为根据实施例所示的仿真地质雷达检测数据；
图11为根据实施例所示的仿真介电常数分布图；
图12为根据实施例所示的仿真目标类别标签；
图13为根据实施例所示的任意长度连续测线地质雷达智能反演与识别预测的介电常数分布图；
图14为根据实施例所示的任意长度连续测线地质雷达智能反演与识别网络预测的目标类别标签。
其中，1.车体，2.移动轮，3.电磁敲击装置，301.敲击锤，302.自动敲击装置外壳，303.安装板，304.第一电磁铁，305.第二电磁铁，306.第一弹性装置，307.第二弹性装置，4.测距编码器，5.声音接收器，6.中间移动轮连接杆，7.驱动电机，8.刚柔弧形杆，9.力矩传感器，10.旋翼助推装置，11.旋翼电机，12.旋翼，13.地质雷达，14.滚珠式超声探头，1401.万向滚珠，1402.软质耦合块，1403.超声探头，1404.弹性装置，1405.柱状外壳，1406.盖子，1407.滚珠，15.移动式超声探头安装框架，16.压力传感器，17.第三弹性装置，18.负压吸附电机，19.负压吸附装置，20.气压传感器，21.多目全景相机，22.激光雷达，23.旋转伸缩装置。
具体实施方式
下面结合附图与实施例对本发明作进一步说明。
应该指出，以下详细说明都是例示性的，旨在对本发明提供进一步的说明。除非另有指明，本文使用的所有技术和科学术语具有与本发明所属技术领域的普通技术人员通常理解的相同含义。
需要注意的是，这里所使用的术语仅是为了描述具体实施方式，而非意图限制根据本发明的示例性实施方式。如在这里所使用的，除非上下文另外明确指出，否则单数形式也意图包括复数形式，此外，还应当理解的是，当在本说明书中使用术语“包含”和/或“包括”时，其指明存在特征、步骤、操作、器件、组件和/或它们的组合。
如图1-图3所示，本实施例的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，包括机器人车体；导航定位系统和移动系统，均安装于机器人车体上；自动敲击检测系统，其安装于车体前端；移动式超声快速检测系统和锈蚀检测系统，均安装于车体底部；双动力系统， 由非接触式负压吸附装置和旋翼助推装置组成，非接触式负压吸附装置安装于车体底部，旋翼助推装置安装于车体两侧；总控制器，其与导航定位系统、移动系统、自动敲击检测系统、移动式超声快速检测系统、锈蚀检测系统和双动力系统分别相互通信。
如图5所示，移动式超声快速检测系统采用滚珠式超声探头阵列，实现超声触点的浮动耦合与移动式检测，探头柱状外壳1405内部一端的球形腔室中装有万向滚珠1401，万向滚珠采用金属、陶瓷等透声性能好的材料，球形腔室内壁装有滚珠1407，可提高万向滚珠滚动流畅性，采用软质耦合块1402连接万向滚珠与超声探头，所述软质耦合块一端与超声探头1403声发射端(或接收端)粘合，另一端具有球面凹槽，用于与万向滚珠贴合，在柱状外壳内部加入超声耦合液，用于填充软质耦合块与万向滚珠之间的空隙，超声探头另一端与弹性装置1404连接，用于辅助耦合。探头柱状外壳上部还设置有盖子1406。
此处需要说明的是，软质耦合块可采用聚胶脂软性塑料、氯丁橡胶、浇注型聚氨酯橡胶等声衰减系数小的材料，本领域技术人员可根据实际情况具体选择。
具体地，若干滚珠式超声探头14安装于机器人内部后侧的移动式超声探头安装框架15，组成超声阵列，滚珠式超声探头上部连接压力传感器16，用于监测实时贴合力，压力传感器另一端连接第三弹性装置17，用于提供贴合力，弹性装置另一端装有旋转伸缩装置23，用于调节贴合力，总控制器根据压力传感器信息调节旋转伸缩装置23，实现各滚珠式超声探头14贴合力相同且稳定，通过超声检测控制器控制每个超声探头的超声信号的发射与接收，接收的信息输入总控制器进行处理。
如图4所示，自动敲击检测系统采用电磁敲击装置3，所述装置包括用于敲击的敲击锤301，敲击锤另一端与第一电磁铁304相连安装于敲击装置外壳内部，第二电磁铁305固定于敲击装置外壳302上部，所述两电磁铁受自动敲击控制器控制与内部的第一弹性装置306和第二弹性装置307配合，实现自动敲击，电磁敲击装置和测距编码器4安装于机器人前部中间移动轮，测距编码器4用于定位，声音接收器5安装于自动敲击装置后侧贴近地面的位置，所述自动敲击检测系统完成缺陷区域快速检测。敲击装置外壳302设置有安装板303。
在具体实施中，锈蚀检测系统采用一种基于地质雷达的两步交错检测模式观测混凝土中钢筋的锈蚀状态，地质雷达13安装于机器人车体底部后侧，第一步为粗测，通过在任意方向行走，获取雷达测线剖面图(B-Scan)，采用基于深度神经网络的钢筋快速定位算法，确定钢筋排布方向；第二部进行锈蚀诊断，沿钢筋布设方向进行扫描，以不同位置的时域信号(A-Scan)振幅等特征参量变化，判别当前位置的锈蚀程度与锈蚀位置。
在本实施例中，双动力系统包含非接触式负压吸附装置19和旋翼助推装置10，所述非接触式负压吸附装置安装于机器人车体内部中间部位且配有气压传感器20，所述旋翼助推装 置安装于机器人车体两侧，力矩传感器9安装于移动系统的刚柔弧形杆8连接车体的一端，用于监测实时贴合力，负压吸附装置与旋翼助推装置均与总控制器连接，总控制器根据力矩传感器9和气压传感器20信息控制是否启动旋翼助推装置，当非接触负压吸附装置所提供的气压达到上限，但仍不满足贴合力需求时，开启旋翼助推装置，实现动力补充。旋翼助推装置10包括旋翼电机11和旋翼12，旋翼电机11用于带动旋翼12旋转。非接触负压吸附装置由负压吸附电机18提供动力。
具体地，移动系统包括刚柔弧形杆、车轮、驱动电机和力矩传感器。采用刚柔弧形杆8连接机器人车体1与车轮，用于减震和传递贴合力，驱动电机7与移动轮2连接，实现机器人的移动和转向，连接车体的一端安装有力矩传感器9，用于监测机器人的实时贴合力。中间移动轮连接杆6连接在机器人前部两个移动轮2之间。
在本实施例中，所述导航定位系统包括多目全景相机21和激光雷达22，采用多目全景相机检测涵闸表面病害，采用激光雷达实现涵闸检测区域自动构图。
本实施例的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作原理为采用“快速普查→精细检测”的检测模式，其具体步骤包括：
步骤(1)：启动机器人双动力系统，使机器人吸附于被侧面，双动力系统在检测过程中实时监测机器人的贴合力，当非接触式负压吸附装置19所提供的贴合力满足需求时，不需开启旋翼助推装置10，反之，需开启旋翼助推装置，实现贴合力补充；
步骤(2)：导航定位系统完成路径规划，进入快速普查模式，启动自动敲击检测系统和锈蚀检测系统，移动系统以高速度移动，电磁敲击装置3以固定频率敲击被侧面，由声音接收器5接收声音信号，同时雷达发射和接收电磁波，微型多目全景相机21拍摄检测区域表面，实现对缺陷区域的快速普查，接收的信息由总控制器进行信息处理，完成缺陷的粗略识别和定位；
步骤(3)：根据快速普查模式所获得的缺陷定位信息，重新进行路径规划，进入精细检测模式，爬壁机器人移动系统以高速移动至缺陷位置，到达缺陷位置后，根据缺陷的类型选择性开启移动式超声快速检测系统和锈蚀检测系统，同时移动系统改为低速移动，移动式超声快速检测系统运行时，总控制器根据压力传感器16信息控制旋转伸缩装置23，用于保持各滚珠式超声探头的贴合力相同且稳定，每完成一处缺陷检测爬壁机器人均以高速移动至另一缺陷位置，检测过程中低速移动，待所有缺陷检测完毕，爬壁机器人返回原点，完成对缺陷区域的精细检测；
基于超声获取的数据采用基于深度学习的识别方法进行处理，采用基于可变形卷积旋转区域检测深度神经网络识别内部缺陷位置与类别，采用融合时序信息的“反演-识别”多任务 深度神经网络从多组超声探测数据中对内部病害连续成像。
不同混凝土结构缺陷形态复杂多样，然而由于衍射、反射、散射等影响，超声图像中的缺陷响应与混凝土缺陷实际形状往往不能一一对应。为了从超声图像中准确识别混凝土缺陷的响应和同相轴形态，本实施例构建了基于可变形卷积的混凝土缺陷响应与位置智能识别方法，引入了可根据缺陷形态任意变化形状的可变形卷积代替传统的规则形状卷积，自适应的提取不同缺陷的同相轴的形态，从而发现并定位任意方向分布的缺陷。
针对移动测量的特点，设计融合时序信息的“反演-识别”多任务深度神经网络，解决了长测线下的内部混凝土缺陷的连续成像问题。与常规的仅基于卷积的编码-解码网络不同，本方法设计了时空特征提取模块以融合不同时刻测量数据的时空信息。首先将多个时刻的测试数据并联输入编码解码网络，其次，重点将解码后的特征并联输入基于ConvLSTM的时空特征提取模块，挖掘前后时刻超声数据的相关性，记忆并融合上一时刻测量结果中涵盖波速、界面的高维特征信息，准确重建连续的缺陷界面与形态。在此基础上，设计多任务分支输出，同时输出缺陷的类型和波速分布，实现了缺陷识别与波速反演一体化。
在另一实施例中，用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，还包括同时反演混凝土介电常数及识别病害，如图6所示，其具体过程为：
步骤S1：建立仿真训练数据集。
针对隧道衬砌病害结构检测问题，建立相应仿真数据集。所述步骤S1具体包括：
步骤S101：构建多种长度的隧道衬砌结构介电常数分布图和目标类别标签。
具体地，对背景介质、病害内部介质、病害数量及病害位置等随机组合，根据每一种组合方式均生成一幅衬砌结构剖面的介电常数分布图。对于每一幅剖面介电常数分布图，根据其包含的各目标体类别，生成与剖面介电常数分布图对应的一副目标类别标签。
其中，隧道衬砌结构模型包含长度为5m、10m、15m等长度的衬砌模型，以模拟出符合实际的多种连续测线长度。
所述背景介质包括素混凝土、钢筋混凝土等多种背景介质，所述病害包括空洞、不密实、裂缝、脱空、断层、溶洞等，病害内部介质为水、空气、泥、岩石等介质。
所述目标体类别依据介电常数模型中的目标进一步分为：钢筋、背景、无水空洞、无水不密实、无水裂缝、无水脱空、无水断层、无水溶洞、含水空洞、含水不密实、含水裂缝、含水脱空、含水断层、含水溶洞等，以更好地识别隧道衬砌内目标体类别。
步骤S102：采用与实际地质雷达子波频率和相位一致的雷克子波作为仿真数据建模的源子波，来对每一介电常数分布图进行正演，生成相应的地质雷达剖面图，并采用直达波去除、增益调整、噪声添加等方法对其进行预处理，来提高真实数据的适应性。
其中，所述正演采用FDTD方法。
所述直达波去除是用获得的仿真地质雷达数据减去其没有病害波形所在道的数据；所述增益调整是通过对比生成仿真地质雷达数据所用的雷克子波和地质雷达设备对着空气采集的空采直达波的幅值水平，对仿真数据的增益进行调整，使其处于同一维度；所述噪声添加是通过对不同现场的背景介质的地质雷达数据进行采集，并从0-1之间随机选择不同的比例将其添加到仿真地质雷达数据中。
步骤S103：通过获得的相应处理后的地质雷达剖面图(如图10)，相应的介电常数分布图(如图11)和目标类别标签(如图12)，得到多组“地质雷达剖面图-介电常数分布图&目标类别标签”数据对，将每组数据对中的介电常数分布图和目标类别标签数据作为地质雷达剖面图的标签，建立仿真训练数据集。
步骤S2：构建并训练任意长度连续测线地质雷达智能反演与识别网络模型
所述步骤S2具体包括：
步骤S201：构建任意长度连续测线地质雷达智能反演与识别网络结构
所述任意长度连续测线地质雷达智能反演与识别网络结构采用“局部短测线空间特征提取结构→长测线时空特征交互结构→介电常数反演与目标识别双任务结构”相级联的实现方式。如图7所示，所述任意长度连续测线地质雷达智能反演与识别网络结构以任意长度连续测线地质雷达检测数据D n(n∈[1,N](N是数据集中全部连续测线地质雷达检测数据的数量)为输入，首先采用步长为S 1的滑动窗口将D n顺序切割为局部短测线B-Scan剖面序列
Figure PCTCN2021111595-appb-000001
(K为D n被切割后的固定大小的局部短测线B-Scan剖面个数)。然后，局部短测线空间特征提取结构并行地对每个局部短测线B-Scan剖面
Figure PCTCN2021111595-appb-000002
进行空间上下文特征提取，得到特征序列
Figure PCTCN2021111595-appb-000003
接着，长测线时空特征交互结构以
Figure PCTCN2021111595-appb-000004
为输入，通过特征序列内的双向时空信息融合，提取包含具有相对稳定的目标特征(包括形状、类别和介电常数)的增强特征序列
Figure PCTCN2021111595-appb-000005
然后，介电常数反演与目标识别双任务结构并行处理每一个局部短测线B-Scan剖面
Figure PCTCN2021111595-appb-000006
的增强特征
Figure PCTCN2021111595-appb-000007
重建对应的介电常数分布图
Figure PCTCN2021111595-appb-000008
和识别目标类别标签
Figure PCTCN2021111595-appb-000009
最后，通过步长为S 2的滑动窗口将
Figure PCTCN2021111595-appb-000010
和
Figure PCTCN2021111595-appb-000011
顺序拼接在一起，重叠位置取平均值，以获得整条连续测线的介电常数图P n和目标类别标签I n。
具体包括三个结构：
(1)局部短测线空间特征提取结构，采用多个并行的全卷积网络结构实现。每个全卷积网络结构单独处理从任意长度连续测线中不同位置提取的各个局部短测线B-Scan剖面
Figure PCTCN2021111595-appb-000012
对其进行空间上下文特征提取，形成任意长度连续测线地质雷达检测数据的特征序列
Figure PCTCN2021111595-appb-000013
作为一种实现方式，所述全卷积网络结构利用DenseUnet网络实现。如图8所示，所述DenseUnet结构由一个一个编码路径和一个解码路径组成，编码路径中包含4个卷积层、4个最大池化层和4个密集块结构，解码路径包含4个卷积层、4个转置卷积层和4个密集块结构，编解码路径之间采用一个卷积层连接。所述编码路径通过逐步下采样将地质雷达检测数据压缩为高级抽象特征，解码路径通过逐步上采样操作获得隧道衬砌内部空间结构的初步表征。所述卷积层结构卷积核大小为3*3，步长为1，通道数为64，以提取地质雷达探测数据中的空间特征。编码路径每个密集块后连接最大池化层，最大池化层内核大小为2*2，步长为2，下采样后的输出被送入下一个编码路径密集块。解码路径的每个转置卷积层连接密集块前，转置卷积过滤器大小为4*4大小，步长为2，通道数为64，。上采样后的输出与来自相应编码路径密集块进行通道维度的拼接，然后送入下一个解码路径密集块。所述密集块结构包含3个串联的卷积层结构，卷积层卷积核大小为3*3，步长为1，通道数为64，每个卷积层将之前所有卷积层的输出信息添加到自身输出，用于促进梯度的流动，更好的学习地质雷达数据的表征。同时，编码路径和解码路径中相应的密集块结构间采用跳层连接，可以将编码过程中的空间信息传递到隧道衬砌内部结构的解码过程。
所述局部短测线空间特征提取结构对任意长度连续测线中提取的不同位置的各个局部短测线B-Scan剖面数据进行空间上下文特征提取，形成地质雷达剖面数据至混凝土内部结构空间信息的初步映射。
(2)长测线时空特征交互结构，采用一层的Bi-ConvLSTM结构实现。所述Bi-ConvLSTM实现局部短测线B-Scan剖面特征序列
Figure PCTCN2021111595-appb-000014
中的双向时空信息融合，通过前向(地质雷达探测方向的正向)和后向(地质雷达探测方向的反向)局部短测线B-Scan特征的自适应信息融合，生成包含局部短测线B-Scan剖面之间的空间上下文信息的增强特征序列
Figure PCTCN2021111595-appb-000015
以用于反演或识别具有连续形状、类别和介电常数值的隧道衬砌内部目标。
作为一种实现方式，如图9所示，所述长测线时空特征交互结构包括一层Bi-ConvLSTM结构。所述Bi-ConvLSTM包含一个前向ConvLSTM层，一个反向ConvLSTM层以及一个卷积层。 所述Bi-ConvLSTM以局部短测线空间特征提取结构所提取的地质雷达数据特征序列
Figure PCTCN2021111595-appb-000016
为输入，前向ConvLSTM层在每个局部短测线B-Scan剖面之间沿着地质雷达探测方向向前传递隐藏状态
Figure PCTCN2021111595-appb-000017
和记忆细胞
Figure PCTCN2021111595-appb-000018
后向ConvLSTM层在每个局部短测线B-Scan剖面之间沿着地质雷达探测方向的反向向后传递隐藏状态
Figure PCTCN2021111595-appb-000019
和记忆细胞
Figure PCTCN2021111595-appb-000020
对于每一个输入的局部短测线剖面特征
Figure PCTCN2021111595-appb-000021
Bi-ConvLSTM单元将对应的前向隐藏状态
Figure PCTCN2021111595-appb-000022
和后向隐藏状态
Figure PCTCN2021111595-appb-000023
在通道维度上连接，然后通过一个卷积层得到最终的包含局部短测线B-Scan剖面空间上下文信息的增强特征。所述ConvLSTM内卷积结构及卷积层卷积结构均采用3*3大小卷积核，步长为1，通道数为64。
所述长测线时空特征交互结构融合任意长度连续测线中提取的不同位置的局部短测线B-Scan数据特征之间的双向时空信息,保证连续测线隧道衬砌结构介电常数反演结果和目标体识别结果的连续性与准确性。
(2)介电常数反演与目标识别双任务结构，采用多个并行的两分支卷积网络结构实现，每个并行的两分支卷积网络结构单独处理每个局部短测线B-Scan剖面
Figure PCTCN2021111595-appb-000024
经长测线时空特征交互结构提取的增强特征
Figure PCTCN2021111595-appb-000025
同时重建隧道衬砌结构介电常数分布和识别病害类型、位置及轮廓。
作为一种实现方式，所述两分支卷积网络结构宝包括反演任务分支和识别任务分支，每个分支包括1个卷积层和1个激活函数。其中，反演任务分支首先采用1层卷积层以降低输入特征维度，卷积核大小为1*1，步长为1，通道数为1。识别任务分支首先采用1层卷积层以降低输入特征维度，卷积核大小为1*1，步长为1，通道数为9。最后，在反演分支的尾部连接一个sigmoid激活函数以回归介电常数分布图，识别分支连接一个softmax函数以分类目标类别。
所述介电常数反演与目标识别双任务结构同时重建混凝土结构介电常数分布和识别病害类型、位置及轮廓，实现地质雷达反演任务和识别任务的特征共享，增强模型泛化能力。
步骤S202：基于仿真训练数据集，训练任意长度连续测线地质雷达智能反演与识别网络模型
基于仿真训练数据集，采用均方误差(MAE)、结构相似性指数(SSIM)与Lovasz_Softmax相结合的损失函数，利用ADAM优化算法对任意长度连续测线地质雷达智能反演与识别网络模型进行优化，训练得到任意长度连续测线地质雷达智能反演与识别网络模型。
步骤S3：建立真实训练数据集，并采用迁移学习方法得到适用于地质雷达实际探测数据的任意长度连续测线地质雷达智能反演与识别网络模型。
所述步骤S3具体包括：
步骤S301：建立真实训练数据集
对地质雷达实际探测的连续测线地质雷达剖面数据、依据实际现场建立的介电常数模型以及目标类型标签进行对应的水平方向的随机裁剪和双线性插值以进行数据增强，建立真实训练数据集。
步骤S302：基于真实训练数据集，采用迁移学习方法得到适用于地质雷达实际探测数据的任意长度连续测线地质雷达智能反演与识别网络模型。
利用真实训练数据集微调基于仿真数据训练的智能反演与识别训练网络模型，使得任意长度连续测线智能反演与识别网络模型学习到真实地质雷达检测数据分布，能够更准确的识别真实衬砌结构的病害。
步骤S4：利用任意长度连续测线地质雷达智能反演与识别网络模型对实际采集到的连续测线地质雷达检测数据进行介电常数反演和病害识别，得到相应的介电常数分布图和目标类型标签。
将训练好的适用于实际数据的智能反演与识别模型参数代入到构建的智能反演和识别网络中，即可得到可以进行实际应用的预测模型。然后，利用Pyqt界面开发工具进行了图形界面的开发，生成可供用户使用的界面，用户可以任意选择采集到的连续测线地质雷达检测数据输入图形界面，然后所述预测模型就会对所述地质雷达检测数据进行反演和识别，生成介电常数分布图和目标类别标签，如图13和图14所示，生成的介电常数分布图和目标类别的存储位置可以由用户自行选择。
根据介电常数分布图和目标类别标签能够还原连续测线测得的衬砌结构的背景介质、病害形态、病害中的填充介质以及病害类别，从而达到病害检测的目的。
当然，上述实施例中，参数的设计、网络的架构等都可以根据具体工况、场景进行更改，这是本领域技术人员容易想到的，理应属于本发明的保护范围，在此不再赘述。
以上所述仅为本公开的优选实施例而已，并不用于限制本公开，对于本领域的技术人员来说，本公开可以有各种更改和变化。凡在本公开的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本公开的保护范围之内。

Claims (18)

  1. 一种用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，包括：
    机器人车体；
    导航定位系统和移动系统，均安装于机器人车体上；
    自动敲击检测系统，其安装于车体前端；
    移动式超声快速检测系统和锈蚀检测系统，均安装于车体底部；
    双动力系统，由非接触式负压吸附装置和旋翼助推装置组成，非接触式负压吸附装置安装于车体底部，旋翼助推装置安装于车体两侧；
    总控制器，其与导航定位系统、移动系统、自动敲击检测系统、移动式超声快速检测系统、锈蚀检测系统和双动力系统分别相互通信。
  2. 如权利要求1所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述导航定位系统包括多目全景相机和激光雷达，多目全景相机用于检测涵闸表面病害，激光雷达用于实现涵闸检测区域的自动构图。
  3. 如权利要求1所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述移动系统包括刚柔弧形杆、车轮、驱动电机和力矩传感器；所述刚柔弧形杆连接机器人车体与车轮，用于减震和传递贴合力；驱动电机与车轮连接，以实现机器人的移动和转向；力矩传感器连接在机器人车体的一端，用于监测机器人的实时贴合力。
  4. 如权利要求1所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述自动敲击检测系统为电磁敲击装置，所述电磁敲击装置包括用于敲击的敲击锤，敲击锤另一端与第一电磁铁相连，第一电磁铁和敲击锤安装于敲击装置外壳内部，第二电磁铁固定于敲击装置外壳上部，两个电磁铁受自动敲击控制器的控制并与敲击装置外壳内部的弹性装置配合，实现自动敲击。
  5. 如权利要求4所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述自动敲击检测系统还包括声音接收器，所述声音接收器安装在电磁敲击装置后侧贴近地面的位置，声音接收器与自动敲击控制器相连，自动敲击控制器与总控制器相连。
  6. 如权利要求1所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述移动式超声快速检测系统采用滚珠式超声探头阵列，每个滚珠式超声探头包括探头和万向滚珠，万向滚珠设置在球形腔室中，球形腔室设置在探头柱状外壳内部，万向滚珠与超声探头采用软质耦合块连接，所述软质耦合块一端与超声探头声发射端或接收端粘合，另一端具有球面凹槽，用于与万向滚珠贴合，柱状外壳内部设有超声耦合液，用于填充软质耦合块与万向滚珠之间的空隙，超声探头另一端与弹性装置连接，用于辅助耦合。
  7. 如权利要求6所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于， 滚珠式超声探头上部连接压力传感器，用于监测实时贴合力，压力传感器另一端连接弹性装置，用于提供贴合力；弹性装置另一端装有旋转伸缩装置，用于调节贴合力；总控制器用于根据压力传感器信息调节旋转伸缩装置，实现各滚珠式超声探头贴合力相同且稳定；每个超声探头与超声检测控制器相连，超声检测控制器与总控制器相连。
  8. 如权利要求1所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述锈蚀检测系统为基于地质雷达的两步交错检测模式观测混凝土中钢筋的锈蚀状态系统，该系统被配置为：通过在任意方向行走，获取雷达测线剖面图，采用基于深度神经网络的钢筋快速定位算法，确定钢筋排布方向；沿钢筋布设方向进行扫描，以不同位置的特征参量变化，判别当前位置的锈蚀程度与锈蚀位置。
  9. 如权利要求3所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统，其特征在于，所述非接触式负压吸附装置配有气压传感器，总控制器用于根据力矩传感器和气压传感器信息控制是否启动旋翼助推装置，当非接触负压吸附装置所提供的气压达到上限，但仍不满足贴合力需求时，开启旋翼助推装置，实现动力补充。
  10. 一种如权利要求1-9中任一项所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，包括：
    启动双动力系统，使机器人吸附于被侧面，双动力系统在检测过程中实时监测机器人的贴合力；
    导航定位系统完成路径规划，进入快速普查模式，启动自动敲击检测系统和锈蚀检测系统，移动系统以高速度移动，利用自动敲击检测系统、锈蚀检测系统、多目全景相机和导航定位系统，实现对缺陷区域的快速普查以及缺陷的粗略识别和定位；
    根据快速普查模式所获得的缺陷定位信息，重新进行路径规划，进入精细检测模式，爬壁机器人移动系统以高速移动至缺陷位置，到达缺陷位置后，根据缺陷的类型选择性开启移动式超声快速检测系统和锈蚀检测系统，同时移动系统改为低速移动，移动式超声快速检测系统运行时，总控制器根据压力传感器信息控制旋转伸缩装置，用于保持各滚珠式超声探头的贴合力相同且稳定，每完成一处缺陷检测爬壁机器人均以高速移动至另一缺陷位置，检测过程中低速移动，待所有缺陷检测完毕，爬壁机器人返回原点，完成对缺陷区域的精细检测；
    基于超声获取的数据采用基于深度学习的识别方法进行处理，采用基于可变形卷积旋转区域检测深度神经网络识别内部缺陷位置与类别，采用融合时序信息的“反演-识别”多任务深度神经网络从多组超声探测数据中对内部病害连续成像。
  11. 如权利要求10所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，还包括同时反演混凝土介电常数及识别病害，其具体过程为：
    对不同应用场景的地质雷达设备提取实际发射子波信号，将其作为仿真建模源子波信号进行正演，生成相应的地质雷达剖面图，其同建立的介电常数分布图和目标类别标签形成数据对，构成仿真训练数据集；
    构建地质雷达智能反演与识别模型，基于仿真训练数据集，训练所述地质雷达智能反演与识别模型；
    建立真实训练数据集，基于真实训练数据集和迁移学习方法微调所述地质雷达智能反演与识别模型；
    利用微调后的地质雷达智能反演与识别模型对真实雷达检测数据进行介电常数反演和病害识别。
  12. 如权利要求11所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，对不同应用场景的地质雷达设备提取实际发射子波信号，将其作为仿真建模源子波信号进行正演，进行正演的具体过程包括：对不同应用场景确定地质雷达设备，提取该地质雷达设备的空采直达波作为子波，采用与该地质雷达设备子波频率和相位一致的雷克子波作为建模的源子波，每一副介电常数分布图进行正演，生成相应的地质雷达剖面图。
  13. 如权利要求11所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，地质雷达剖面图同建立的介电常数分布图和目标类别标签形成数据对的具体过程包括：将不同应用场景测量所得的背景噪声随机叠加到仿真雷达检测数据中，得到背景与子波接近真实的仿真数据；建立混凝土结构介电常数分布图和目标类别标签两种标签，得到“地质雷达剖面图-介电常数分布图与/或目标类别标签”数据对。
  14. 如权利要求11所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，所述地质雷达智能反演与识别模型包括级联的局部短测线空间特征提取结构、长测线时空特征交互结构和介电常数反演与目标识别双任务结构。
  15. 如权利要求14所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，所述局部短测线空间特征提取结构，采用多个并行的全卷积网络结构，每个全卷积网络结构用于单独处理从任意长度连续测线中不同位置提取的各个局部短测线B-Scan剖面，对其进行空间上下文特征提取，形成任意长度连续测线地质雷达检测数据的特征序列。
  16. 如权利要求14所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，所述长测线时空特征交互结构，采用一层Bi-ConvLSTM结构，用于实现局部短测线B-Scan剖面特征序列中的双向时空信息融合，通过地质雷达探测方向的正向和地质雷达探测方向的反向的局部短测线B-Scan特征的自适应信息融合，生成包含局部短测线 B-Scan剖面之间的空间上下文信息的增强特征序列，以实现反演或识别具有连续形状、类别和介电常数值的混凝土内部目标。
  17. 如权利要求14所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，所述介电常数反演与目标识别双任务结构，采用多个并行的两分支卷积网络结构，每个并行的两分支卷积网络结构用于单独处理每个局部短测线B-Scan剖面经长测线时空特征交互结构提取的增强特征，同时重建混凝土介电常数分布和识别病害类型、位置及轮廓。
  18. 如权利要求11所述的用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统的工作方法，其特征在于，建立真实训练数据集的具体过程包括：对地质雷达实际探测的连续测线地质雷达剖面数据、依据实际现场建立的介电常数模型以及目标类型标签进行对应的水平方向的随机裁剪和双线性插值以进行数据增强，建立真实训练数据集。
PCT/CN2021/111595 2021-01-19 2021-08-09 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法 Ceased WO2022156192A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/765,215 US12072298B2 (en) 2021-01-19 2021-08-09 Wall-climbing robot system and method for rapid nondestructive inspection of hidden defects in culverts and sluices

Applications Claiming Priority (4)

Application Number Priority Date Filing Date Title
CN202110071015.7A CN112917483B (zh) 2021-01-19 2021-01-19 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法
CN202110071015.7 2021-01-19
CN202110706142.XA CN113447536B (zh) 2021-06-24 2021-06-24 一种混凝土介电常数反演与病害识别方法及系统
CN202110706142.X 2021-06-24

Publications (1)

Publication Number Publication Date
WO2022156192A1 true WO2022156192A1 (zh) 2022-07-28

Family

ID=82549356

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/111595 Ceased WO2022156192A1 (zh) 2021-01-19 2021-08-09 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法

Country Status (2)

Country Link
US (1) US12072298B2 (zh)
WO (1) WO2022156192A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115588166A (zh) * 2022-11-10 2023-01-10 新乡市诚德能源科技装备有限公司 一种防泄漏船用lng燃料罐
CN116203557A (zh) * 2023-03-06 2023-06-02 北京交通大学 一种基于探地雷达的传统石砌墙体内部残损及异常物无损辨识方法
CN117446042A (zh) * 2023-10-25 2024-01-26 广州高新兴机器人有限公司 核电站监测机器人及监测方法
CN119087541A (zh) * 2024-11-01 2024-12-06 山东大学 基于多模态学习的隧道震-电-雷达联合反演与解析方法
NL2039276A (en) * 2023-12-19 2025-01-09 Univ Chongqing Jiaotong A method and device for detecting voids in tunnel lining structures
CN120102574A (zh) * 2025-04-27 2025-06-06 宁德思客琦智能装备有限公司 一种激光焊接质量监控方法及系统
CN120600192A (zh) * 2025-08-05 2025-09-05 水利部交通运输部国家能源局南京水利科学研究院 长距离地下涵洞结构健康评估方法

Families Citing this family (14)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117307990A (zh) * 2023-10-27 2023-12-29 合肥众安睿博智能科技有限公司 基于辅助传感协同惯性导航的封闭场景运动轨迹拟合技术
DE102024202234A1 (de) 2024-03-11 2025-09-11 Robert Bosch Gesellschaft mit beschränkter Haftung Verfahren zum Erzeugen eines Trainingsdatensatzes
WO2025235465A1 (en) * 2024-05-06 2025-11-13 Quest Integrity Usa, Llc. Methods and systems for tank inspections
CN118481025B (zh) * 2024-07-15 2024-10-25 江西省天驰高速科技发展有限公司 一种矩形墩柱检测机器人
CN119305647B (zh) * 2024-10-17 2025-11-11 青岛中远海运数智科技有限公司 一种自适应变曲率曲面装置检测的轮式爬壁机器人
CN119370224B (zh) * 2024-10-28 2025-11-28 青岛中远海运数智科技有限公司 一种爬壁机器人的行走机构及焊缝检修机器人
CN119321919B (zh) * 2024-12-18 2025-04-11 大连三丰混凝土预制构件有限公司 墙体混凝土取芯装置及混凝土检测系统
CN119414004B (zh) * 2025-01-09 2025-04-15 天津理工大学 一种工程质量检测装置及其检测方法
CN119666986B (zh) * 2025-02-19 2025-07-22 杭州海康威视数字技术股份有限公司 缺陷检测方法、装置及电子设备
CN119919418B (zh) * 2025-04-03 2025-08-26 水利部珠江水利委员会技术咨询(广州)有限公司 一种基于深度学习图像处理的堤防表面缺陷检测方法
CN120009188B (zh) * 2025-04-18 2025-07-15 四川航空股份有限公司 一种可以识别缺陷标记的民航维修用外舱攀爬探伤设备
CN120446287B (zh) * 2025-05-06 2026-03-06 盱眙国联建设工程质量检测有限公司 一种基于超声波技术的道路无损检测方法及系统
US12578496B1 (en) 2025-08-29 2026-03-17 Smartauger, Inc. B-scan normalization for non-destructive testing
US12510516B1 (en) * 2025-09-17 2025-12-30 Prince Mohammad Bin Fahd University Modular pipe inspection robot and method for inspecting an external surface of a pipe

Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN205554355U (zh) * 2016-04-05 2016-09-07 浙江科技学院 空鼓探测爬墙机器人
CN207207653U (zh) * 2017-09-14 2018-04-10 王志成 基于四旋翼结构的爬墙机器人
CN109795571A (zh) * 2019-01-22 2019-05-24 浙江工业大学 一种用于平面建筑结构探伤的爬壁机器人检测系统
US20190353819A1 (en) * 2018-04-27 2019-11-21 Schlumberger Technology Corporation Determining Dielectric Constant and Resistivity with Induction Measurement
CN209745884U (zh) * 2019-01-22 2019-12-06 国电锅炉压力容器检验有限公司 一种超声波探头
CN110865128A (zh) * 2019-12-18 2020-03-06 上海市建筑科学研究院有限公司 一种自动爬墙式外墙检测系统及其检测方法
CN111751392A (zh) * 2020-07-30 2020-10-09 广州大学 一种基于双极化探地雷达的钢筋锈蚀检测方法
CN111781576A (zh) * 2019-10-09 2020-10-16 山东大学 一种基于深度学习的探地雷达智能反演方法
CN112917483A (zh) * 2021-01-19 2021-06-08 山东大学 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11420692B2 (en) 2016-03-18 2022-08-23 Invert Robotics Limited Surface wave detection of surface defects
CN105891454A (zh) 2016-05-23 2016-08-24 桂仲成 轮毂式道面自主检测机器人系统及检测方法
CN108731736B (zh) 2018-06-04 2019-06-14 山东大学 用于桥隧结构病害无损检测诊断的自动爬墙式雷达光电机器人系统
CN109324112B (zh) 2018-10-09 2020-11-10 河海大学 一种基于巨磁电阻元件的闸门检测机器人及检测方法
CN109324121B (zh) 2018-10-09 2020-07-17 河海大学 一种基于相控阵超声波探伤仪的闸门检测装置及检测方法
CN109459494A (zh) 2018-11-22 2019-03-12 中建四局第六建筑工程有限公司 一种用于钢管内柱高标号山砂混凝土的检测方法及装置
CN109794945A (zh) 2019-01-22 2019-05-24 浙江工业大学 一种用于建筑结构探伤的自动敲击机器人检测系统
US12122467B2 (en) * 2019-05-30 2024-10-22 Tokyo Keiki Inc. Moving device
CN110920875A (zh) 2019-11-15 2020-03-27 武汉理工大学 一种多旋翼增压式飞行爬壁机器人

Patent Citations (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN205554355U (zh) * 2016-04-05 2016-09-07 浙江科技学院 空鼓探测爬墙机器人
CN207207653U (zh) * 2017-09-14 2018-04-10 王志成 基于四旋翼结构的爬墙机器人
US20190353819A1 (en) * 2018-04-27 2019-11-21 Schlumberger Technology Corporation Determining Dielectric Constant and Resistivity with Induction Measurement
CN109795571A (zh) * 2019-01-22 2019-05-24 浙江工业大学 一种用于平面建筑结构探伤的爬壁机器人检测系统
CN209745884U (zh) * 2019-01-22 2019-12-06 国电锅炉压力容器检验有限公司 一种超声波探头
CN111781576A (zh) * 2019-10-09 2020-10-16 山东大学 一种基于深度学习的探地雷达智能反演方法
CN110865128A (zh) * 2019-12-18 2020-03-06 上海市建筑科学研究院有限公司 一种自动爬墙式外墙检测系统及其检测方法
CN111751392A (zh) * 2020-07-30 2020-10-09 广州大学 一种基于双极化探地雷达的钢筋锈蚀检测方法
CN112917483A (zh) * 2021-01-19 2021-06-08 山东大学 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法

Cited By (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115588166A (zh) * 2022-11-10 2023-01-10 新乡市诚德能源科技装备有限公司 一种防泄漏船用lng燃料罐
CN115588166B (zh) * 2022-11-10 2023-02-17 新乡市诚德能源科技装备有限公司 一种防泄漏船用lng燃料罐
CN116203557A (zh) * 2023-03-06 2023-06-02 北京交通大学 一种基于探地雷达的传统石砌墙体内部残损及异常物无损辨识方法
CN116203557B (zh) * 2023-03-06 2024-03-05 北京交通大学 一种基于探地雷达的传统石砌墙体内部残损及异常物无损辨识方法
CN117446042A (zh) * 2023-10-25 2024-01-26 广州高新兴机器人有限公司 核电站监测机器人及监测方法
NL2039276A (en) * 2023-12-19 2025-01-09 Univ Chongqing Jiaotong A method and device for detecting voids in tunnel lining structures
CN119087541A (zh) * 2024-11-01 2024-12-06 山东大学 基于多模态学习的隧道震-电-雷达联合反演与解析方法
CN120102574A (zh) * 2025-04-27 2025-06-06 宁德思客琦智能装备有限公司 一种激光焊接质量监控方法及系统
CN120600192A (zh) * 2025-08-05 2025-09-05 水利部交通运输部国家能源局南京水利科学研究院 长距离地下涵洞结构健康评估方法

Also Published As

Publication number Publication date
US20240068951A1 (en) 2024-02-29
US12072298B2 (en) 2024-08-27

Similar Documents

Publication Publication Date Title
WO2022156192A1 (zh) 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法
CN112917483B (zh) 用于涵闸隐蔽缺陷快速无损检测的爬壁机器人系统及方法
CN113424055B (zh) 隧道结构病害多尺度检测与智能诊断系统及方法
CN113447536B (zh) 一种混凝土介电常数反演与病害识别方法及系统
US11640007B2 (en) System and method for phased array sound wave advanced geological exploration for shield tunneling machine
CN113204824B (zh) 多模型融合的盾构施工沉降预测方法及系统
CN111781576A (zh) 一种基于深度学习的探地雷达智能反演方法
CN111981968B (zh) 基于地质雷达小波变换的隧道衬砌厚度识别方法及系统
CN115749734B (zh) 一种随钻超声成像测井仪在线学习地质构造的方法
CN110824568A (zh) 一种搭载于盾构机刀盘-内置聚焦式电法探水系统及方法
Jin et al. Gpr-former: Detection and parametric reconstruction of hyperbolas in gpr b-scan images with transformers
CN120426065A (zh) 一种基于地层结构的盾构适应性调整方法
CN121069517A (zh) 基于多物探联合反演与多源数据融合的隧洞超前地质预报方法
CN121028077B (zh) 一种基于探地雷达的地下管线分布检测方法及系统
CN212868986U (zh) 一种用于排水管道检测的鲔科仿生鱼机器人
CN110057915B (zh) 渠道工程水下全断面三维基础成像检测方法
CN119917808B (zh) 基于联邦学习的岩溶地层盾构掘进参数预测方法及系统
CN119107236B (zh) 探地雷达B-scan图像超分辨率处理方法及装置
CN109828317B (zh) 一种耦合接收装置、tbm掘进隧洞精细化探测系统及方法
Li et al. DDA-GPR: symmetric adversarial domain adaptation for few-shot underground cavity identification in urban roads
CN119535446A (zh) 一种tbm掘进机掘进时的不良地质体识别方法、系统及设备
Guo et al. Research on tunnel lining image target recognition method based on YOLOv3
Li et al. Intelligent Identification of Pile Defects Based on Improved LSTM Model and Wavelet Packet Local Peaking Method.
Zhang et al. Research on underwater nondestructive testing technology of hydraulic culvert based on ROV
CN121235231B (zh) 基于多源数据融合的煤矿采空区塌陷智能预测系统

Legal Events

Date Code Title Description
WWE Wipo information: entry into national phase

Ref document number: 17765215

Country of ref document: US

121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21920573

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21920573

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 17/01/2024)

122 Ep: pct application non-entry in european phase

Ref document number: 21920573

Country of ref document: EP

Kind code of ref document: A1