A pipeline detection robot based on spiral scanning and detection method thereof

Abstract

The present invention designs a pipeline detection robot based on spiral scanning and a detection method thereof, comprising a crawler body, a spiral detection system, a control mechanism and a sensing mechanism; the spiral detection system comprises a detection arm lifting platform and a cross-type detection arm, the control mechanism adjusts the height of the detection arm lifting platform and drives the cross-type detection arm to perform spiral scanning; the processing and correction of the spiral line signal are completed by a set of spiral signal inversion algorithms; the axis center control method is realized by controlling the height of the detection arm lifting platform through information collection of the sensing mechanism and feedback of the control mechanism; the present invention is designed to solve the problem of defect signal leakage caused by the detection method of the existing PIG-type pipeline robot with dense sensors, the axis center control method ensures the stability of the robot's traveling detection, and the spiral detection method and its signal inversion algorithm can realize all-round scanning of the pipeline without omission.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline detection, and in particular to a pipeline detection robot based on spiral scanning and a detection method thereof. Background Art

[0002] Pipeline transportation plays an important role in industrial development and urban modernization. In the transportation of energy such as oil and natural gas, pipelines are widely used as an effective way of material transportation, which has brought huge economic benefits to industrial development and people's lives and has immeasurable development prospects.

[0003] However, defects may exist in pipelines during manufacturing and operation, and their damage will bring huge losses, waste of resources, environmental pollution, poisoning, explosion and other accidents. Therefore, in order to extend the life of the pipeline and prevent leakage accidents, it is necessary to effectively detect, maintain and repair the pipeline.

[0004] The pipeline environment is complex and long, and manual inspection operations are difficult to perform. Pipeline robots are suitable for pipeline inspection, cleaning, and flaw detection because of their small size, ability to crawl in pipelines, ability to carry a variety of sensors and operating machinery, and remote control. Therefore, in order to ensure the safe operation of oil and gas pipelines, it is necessary to use pipeline inspection robots to conduct in-pipeline inspections.

[0005] PIG robots are the most representative oil and gas pipeline inspection equipment. They do not have autonomous driving capabilities. Their power to move forward in the oil and gas pipeline comes from the pressure difference between the fluid at the head and the end. They do not need to tow cables or carry batteries for driving. They can travel hundreds of kilometers at a time, but their load capacity is low and they cannot move forward autonomously. Instability during movement and changes in the gap between sensor arrangements may lead to errors in data collection.

[0006] Schempf H et al. from Carnegie Mellon University in the United States developed the Explorer series of urban natural gas pipeline inspection robots, among which Explorer I1 can be equipped with far-field eddy current or magnetic flux leakage detectors for pipeline defect detection. Pipe corrosion, mechanical damage and wall thinning can be effectively detected through ultrasonic, far-field eddy current or magnetic flux leakage methods. However, during movement, defect signals may be missed due to excessive gaps between sensors.

[0007] The research on pipeline robots in my country started relatively late. In 1994, China National Petroleum and Natural Gas Pipeline Bureau introduced magnetic flux leakage detection equipment from the United States and began to truly study and apply magnetic flux leakage detection technology. In the process of many years of research and application development, the detection technology level of the pipeline bureau has made a huge leap. The magnetic flux leakage detection technology of the pipeline bureau is basically close to the international level, but there are still problems in the detection of crack defects such as stress cracks and fatigue cracks. At present, although the country has basically mastered the external detection service technology of oil and gas pipelines, it is still vacant in the internal detection service technology of pipelines. Compared with magnetic flux leakage detection, far-field eddy current can detect both internal and external defects, and has a small requirement for lift-off value. The research on electromagnetic detection methods such as far-field eddy current is just getting started and has great development potential. Summary of the invention

[0008] In view of the deficiencies in the prior art, the present invention designs a pipeline inspection robot based on spiral scanning and an inspection method thereof.

[0009] A pipeline inspection robot based on spiral scanning comprises a crawler body, a spiral inspection system, a sensing mechanism and a control mechanism; the spiral inspection system is installed on the rear half of the body, the sensing mechanism is connected to the control mechanism, the control mechanism is connected to the spiral inspection system, and both the sensing mechanism and the control mechanism are connected to the crawler body;

[0010] The crawler vehicle body comprises a vehicle body frame, a driving wheel, a driven wheel, a driving motor, a driving motor fixing frame, a bevel gear set, a ball screw, a multi-connection screw nut, a first screw support plate, a second screw support plate, a driving reducer, a mileage wheel fixing groove, a first driven reducer and a second driven reducer, a first track fixing plate and a second track fixing plate, a track inner baffle plate and a track outer baffle plate, a bearing, and an excitation coil fixing plate; the two driving wheels and the two driven wheels are divided into two groups and symmetrically arranged on both sides of the vehicle body, the two driving wheels are respectively connected to the main gears of the two bevel gear sets, the sub gears of the two bevel gear sets are respectively connected to the two driving motors, and the driving motors The driving motor fixing frame is fixed to the inner fender of the track, the driving wheel and the driven wheel are connected to the inner fender of the track and the outer fender of the track through bearings, the inner fender of the track and the outer fender of the track are respectively connected to the first driven reducing rod and the second driven reducing rod through the first track fixing plate and the second track fixing plate, and are connected to the vehicle body frame through the first driven reducing rod and the second driven reducing rod, the two ends of the ball screw are respectively fixed to the first screw support plate and the second screw support plate, the multi-connection screw nut is sleeved on the ball screw and moves forward and backward along the ball screw, the three active reducing rods are connected to the side of the multi-connection screw nut, and the two active reducing rods located at the bottom are connected to the second track fixing plate;

[0011] An active diameter reducing rod is installed above the ball screw, a mileage wheel fixing groove is installed on the top of the active diameter reducing rod, and an exciting coil fixing plate is fixed on the front end of the first screw support plate;

[0012] The spiral detection system includes a detection arm lifting platform and a cross-type detection mechanism. The detection arm lifting platform includes a lifting platform base, a steering gear, a steering gear gear, a telescopic ruler bar, and a rotating motor bracket; the cross-type detection mechanism includes a rotating motor, a coupling, a center connecting ring, n detection arms, each of which includes a sleeve, a telescopic rod, an elastic structure, and a detection probe. The lifting platform base is fixed on the upper surface of the vehicle frame. The telescopic ruler bar is connected to the lifting platform base and slides relatively. The steering gear contacts the telescopic ruler bar through the steering gear gear. The rotating motor is fixed to the top of the telescopic ruler bar through the rotating motor bracket and moves up and down with it. The central connecting ring of the cross-type detection mechanism is connected to the rotating motor through a coupling. n The detection arms are Angle Distributed outside the central connecting ring and fixed thereto, the sleeve of the detection arm is connected to the telescopic rod through a slot structure, an elastic structure is installed inside the sleeve to change the length of the detection arm, the other end of the telescopic rod is fixed to the detection probe, and the detection probe reserves space for winding the detection coil and wiring;

[0013] The control mechanism includes a controller, a power supply, a motor drive module, a DC step-down module, and a signal transmission module. The power supply is connected to the controller, the motor drive module, and the DC step-down module. The controller is connected to the motor drive module and the signal transmission module through wires, respectively. The DC step-down module and the motor drive module are both connected to the drive motor and the rotating motor through wires. The controller, the power supply, the motor drive module, the DC step-down module, and the signal transmission module are installed on the vehicle body frame.

[0014] The sensing mechanism includes an electronic gyroscope, an infrared ranging module, and a Hall speed measurement module; the electronic gyroscope is installed on the lower surface of the vehicle frame and is parallel to the surface of the crawler vehicle body, the infrared ranging module is fixed on the rotating motor bracket in the detection arm lifting platform, and the Hall speed measurement module is connected to the driving motor and the rotating motor of the crawler vehicle body;

[0015] A detection method of a pipeline detection robot based on spiral scanning is implemented based on the above pipeline detection robot based on spiral scanning, and comprises the following steps:

[0016] Step 1: Set the robot's drive motor initial speed n0, the ratio of the rotation motor speed to the drive motor speed K, the spiral detection system eccentricity threshold TH, and the sampling width h , detect the maximum allowable sampling width error e, where the ratio of the rotating motor speed to the driving motor speed K is a constant value determined by the maximum effective detection range of the detection probe, the eccentricity threshold TH of the spiral detection system is a constant value determined by the diameter range adapted to the small elastic diameter change of the detection arm, and the sampling width h It is a constant value determined by the sampling principle of the detection probe and detects the maximum allowable sampling width error. e It is a constant value determined by the scanning accuracy requirement of the detection probe;

[0017] Step 1.1: Establish a right-handed rectangular coordinate system inside the oil and gas pipeline to be inspected, where the robot movement direction along the central axis of the pipeline is the positive direction of the Z axis. Assuming that the cross-type detection mechanism rotates clockwise when viewed from the front of the robot, the movement of the center point of the detection area of ​​the detection probe on the inner wall of the pipeline is the combined motion of the linear motion along the Z axis and the circumferential circular motion in the OXY plane. For any initial position of the projection of the detection probe of a certain detection arm on the inner wall of the pipeline , its motion trajectory equation is:

[0018]

[0019] In the formula, is the rotational angular velocity of the detection arm in the cross-type detection mechanism, v is the robot's moving speed, t For working hours, x , y , z are respectively the intercepts of the projection position of the detection probe of the detection arm on the inner wall of the pipeline on the XYZ axis;

[0020] Step 1.2: Align the cylindrical surface where the pipe is located along a straight line x = R , y =0 and unfolded into a Cartesian coordinate system. The unfolded image of the inner surface of the pipe scanned by the robot is a rectangle with a length of L , width is W ;

[0021] and L = S , W = C , C =2π R , S = vt ;

[0022] In the formula, S To detect the distance traveled by the vehicle during working hours, C is the circumference of the pipe section, R is the inner radius of the pipe;

[0023] The spiral detection trajectory will be unfolded into an inclined straight line in the Cartesian coordinate system, and the angle between the straight line and the length of the rectangle is for:

[0024]

[0025] The distance between the centers of two adjacent trajectories d for:

[0026]

[0027] In the formula, n is the number of detection arms involved in scanning in the spiral detection system;

[0028] The difference between the distance between the two detection tracks and the sampling width should be less than the maximum allowable sampling width error to avoid scanning omissions and scanning duplications. After calculation, the conditions that need to be met are:

[0029]

[0030] In the formula, e To detect the maximum allowable sampling width error, h is the sampling width;

[0031] Initially set the ratio K of the robot's rotation motor speed to the drive motor speed:

[0032]

[0033] In the formula, r is the radius of the robot's driving wheel;

[0034] In summary, in order to ensure that the detection arm fully scans the inner wall of the pipeline, the control algorithm needs to be used to maintain the ratio K of the rotating motor speed to the driving motor speed to meet the following conditions:

[0035]

[0036] Ideally, when detecting the maximum allowable sampling width error e When it is 0, the ratio K of the rotating motor speed to the driving motor speed is calculated using the following formula:

[0037]

[0038] Step 2: The infrared ranging module of the sensing mechanism obtains the position information of the central connecting ring of the cross-type detection mechanism in the axial direction of the pipeline and feeds it back to the control mechanism. The control mechanism adjusts the rotation angle of the servo to control the height of the detection arm lifting platform. Through the axis center control method, the geometric center of the cross-type detection mechanism is always at the axial center of the pipeline.

[0039] Step 2.1: The infrared distance measuring module on the top of the detection arm lifting platform transmits the collected distance information from the inner wall of the pipeline to the control mechanism in real time;

[0040] Step 2.2: The control mechanism calculates the eccentricity X of the position of the center of the cross-type detection mechanism on the pipeline cross section according to the distance information obtained in real time;

[0041] Step 2.3: Compare the acquired relevant parameters with the set eccentricity threshold TH of the spiral detection system, control the lifting platform through the position PID algorithm, and adjust the rotation direction, angle and speed of the servo;

[0042] Step 2.4: Repeat steps 2.1 to 2.3 until the eccentricity X is always controlled to be less than or equal to the eccentricity threshold TH, and then perform output limiting;

[0043] Step 3: The control mechanism uses an incremental PID feedback algorithm to achieve speed closed-loop control according to the required setting value of the driving motor speed, and drives the driving motor of the crawler body and the rotating motor of the spiral detection system to work respectively. The Hall speed measurement module obtains the rotation speeds of the driving motor of the crawler body and the rotating motor of the spiral detection system, and transmits them to the control mechanism. The control mechanism compares the rotation speeds of the driving motor of the crawler body and the rotating motor of the spiral detection system obtained by the Hall speed measurement module with the initial rotation speed n0 of the driving motor of the robot and the ratio K of the rotating motor speed to the driving motor speed, and performs feedback control to achieve the set rotation speeds of the driving motor and the rotating motor, so as to ensure that the inner wall of the pipeline is scanned without omission during the travel process;

[0044] Step 4: When the robot is working, the detection probe moves in a spiral trajectory close to the inner wall of the pipeline to collect signals. The spiral signal is inverted and corrected through a set of spiral signal inversion algorithms, and the three-dimensional pipeline inner wall information is converted into a two-dimensional pipeline inner wall expansion diagram to realize the visualization of pipeline inner wall defects;

[0045] The spiral signal inversion algorithm inverts and corrects the spiral signal, specifically:

[0046] Step 4.1: Use Matlab to read the sampling data of the detection probe and the vehicle body operation data from the data storage device, including n The voltage value measured by the detection arm, the phase of the detection arm, the time of the trolley movement, and the number of motor shaft rotation pulses measured by the Hall speed measurement module;

[0047] Step 4.2: Preliminarily identify the defect values ​​and abnormal values ​​of the sampled data, filter out the noise signal, and then fit the preliminarily processed data, perform interpolation data preprocessing, and perform approximate processing on the data with more decimal places;

[0048] Step 4.3: Read the preprocessed data into the structure data type to form a data set, and use the new stacking method to process the data according to the actual physical meaning to increase the dimension of the data set, and finally add interactive features to each sampled pipe wall signal, including the phase when measuring the value, time data, and the number of motor shaft rotation pulses;

[0049] Step 4.4: The clock drive is converted into a mileage drive algorithm to obtain the mileage of the car. At the same time, the electronic gyroscope signal on the car body is used to correct the signal of the detection arm to reduce the detection data error caused by the tilt of the car body. After correction, it is inserted into the data set to finally obtain a database of sampled data;

[0050] The specific steps of converting the clock drive into the mileage drive algorithm in the spiral signal inversion algorithm are as follows:

[0051] Step 4.4.1: Extract the data set of the vehicle movement time and the motor shaft rotation pulse number;

[0052] Step 4.4.2: Read the number of encoder pulses per unit clock measured by the Hall speed measurement module and calculate the forward speed of the car;

[0053] Step 4.4.3: Calculate the mileage of the car by using the step-by-step integration algorithm from the car's forward speed;

[0054] Step 4.4.4: Compare and correct the mileage of the car obtained from steps 4.4.1 to 4.4.3 with the mileage data collected by the sensing mechanism to reduce the accidental errors caused by these two methods;

[0055] Step 4.5: Use the loop structure to use the vehicle body inclination information collected by the electronic gyroscope as the characteristic value index, find the voltage values ​​of different spiral arms at different mileage coordinates of the phase in the database in turn, and finally obtain the database of each phase value. After drawing, the pipeline inner wall expansion diagram is obtained to complete the data visualization work;

[0056] Step 5: In order to further judge the type of defect and the specific location of the defect in the pipeline inner wall expansion diagram, a pipeline inner wall defect recognition model based on deep learning is proposed. In the output result of the model, the defect feature information judgment algorithm is used to identify and mark the geometric features of the defect, including the defect area, perimeter, maximum boundary distance, and maximum depth. At the same time, the number of meters and phases of the defect in the pipeline are marked, and finally the pipeline inner wall expansion diagram after processing is output;

[0057] Step 5.1: Obtain data by collecting the pipeline inner wall development diagram obtained in step 4;

[0058] Step 5.2: First, perform preliminary cropping on the original image, and then perform noise filtering and image enhancement preprocessing on the cropped image;

[0059] The cropping is to use a screenshot tool to crop the original image to obtain the target measurement area, and the image is saved in "jpg" format;

[0060] Step 5.3: Train the target detection network model (Faster-RCNN) based on the region proposal algorithm and obtain the pipeline inner wall defect recognition model;

[0061] Step 5.3.1: Initialize each module of the neural network's unique layer parameters using pre-trained weights from a large-scale image recognition database (ImageNet);

[0062] Step 5.3.2: Collect inspection images of normal pipeline inner wall and pipeline inner wall with defects, classify and mark them, determine training set and test set, and form a sample library;

[0063] Step 5.3.2.1: Use the detection probe to detect various known types of defects and the inner wall of the normal pipeline, and generate a detection image;

[0064] Step 5.3.2.2: Identify and label the defects on the inner wall of the pipeline; select the defect images with obvious features, remove the invalid data that are difficult to identify or have been repaired, and segment the selected images using a fixed pixel size. Then use the data labeling tool software Labelme to label the types of all defects. The labels correspond to the defect types one by one to obtain a set of labeling sets;

[0065] Step 5.3.2.3: Determine the training set and test set: According to the number of samples, divide the training samples into training set and test set in a ratio of 7:3, integrate the txt files, pictures and xml files into files that can be read by Faster_RCNN to form a sample library;

[0066] Step 5.3.3: The training is divided into two parts, one is to train the RPN network, and the other is to train the subsequent classification network; use the data set in step 5.3.2 to train the parameters of each network layer in turn;

[0067] Step 5.3.4: After testing the test set, fine-tune the parameters of each network layer according to the test results and the actual error, and finally form a unified network to obtain the pipeline inner wall defect recognition model;

[0068] Step 5.4: Use the model to perform defect identification process, determine whether there are defects on the inner wall of the pipeline, frame the defect location and mark the type;

[0069] Step 5.4.1: Scale the detection image to a fixed size and then pass it into the feature extraction network. The feature extraction network uses a deep residual network to obtain a feature map;

[0070] Step 5.4.2: The obtained feature map is passed to the RPN part to generate a proposed candidate box where defects may occur, determine whether there are defects in the candidate box, and make the range of the candidate box more accurate by correcting the box;

[0071] Step 5.4.3: Extract the defect suggestion candidate box generated in step 5.4.2 and the defect feature map in step 5.4.1, and pass them to the ROI Polling layer to obtain a fixed-size feature map containing the candidate box and send it to the subsequent fully connected layer for classification and regression;

[0072] Step 5.4.4: Send the feature map of the fixed-size candidate box obtained in step 5.4.3 to the subsequent classification layer and regression layer for classification and regression operations to obtain the specific location of the target;

[0073] Step 5.5: Determine the defect feature information including geometric feature information and orientation feature information based on the defect feature information processing algorithm;

[0074] Step 5.5.1: Get the framed defects identified by Faster R-CNN;

[0075] Step 5.5.2: By means of image enhancement, the image is converted into a grayscale image through grayscale transformation;

[0076] Step 5.5.3: Use the median filter method to perform noise filtering on the grayscale image, and then perform morphological processing on the grayscale image after noise filtering, including corrosion and expansion operations;

[0077] Step 5.5.4: Use the Canny edge detection algorithm to determine the defect edge position;

[0078] Step 5.5.5: Determine the target defect in the image, and then obtain the total number of pixels of the target defect. By comparing it with the number of pixels of the standard object, various geometric feature information of the defect is finally obtained, including the defect area, perimeter, and maximum boundary distance.

[0079] Step 5.5.6: Recall the defect information in the database, convert it, and obtain the maximum depth or maximum bulge height of the defect, as well as the position characteristic information of the defect in the pipeline, including the position and phase of the defect in the pipeline, and finally add all the characteristic information to the annotation;

[0080] Step 5.6: Output the pipeline inner wall defect expansion diagram after processing.

[0081] Beneficial technical effects of the present invention:

[0082] Aiming at the problem of missed detection of defect signals of densely distributed PIG-type pipeline inspection robots, the present invention designs a pipeline inspection robot based on spiral scanning and its inspection method. The spiral scanning system applies the spiral inspection concept of nuclear magnetic resonance imaging in the medical field to industrial oil and gas pipeline inspection for the first time, and can realize full-range scanning of the inner wall of the pipeline without omission; the spiral signal inversion algorithm realizes the processing of spiral signals and the location of defects in the pipeline, avoiding the problem of missed detection of defect signals caused by the gap between sensors in the arrangement of traditional structures; the detection system axis center control method always controls the rotation center of the cross-type detection mechanism to the axial center of the pipeline to ensure the stability of the detection. This detection method can ultimately realize the detection of defects such as damage, cracks, and depressions on the pipeline wall within a certain caliber range without omission. BRIEF DESCRIPTION OF THE DRAWINGS

[0083] Figure 1 It is a three-dimensional diagram of the overall structure of the present invention;

[0084] Figure 2 A three-dimensional diagram of the walking structure of the crawler-type vehicle body of the present invention;

[0085] Figure 3 A three-dimensional diagram showing the assembly relationship between the crawler chassis and the spiral detection system of the present invention;

[0086] Figure 4 It is a three-dimensional diagram of the spiral detection system structure in the present invention;

[0087] Figure 5 This is a flow chart of the central axis center control method of the present invention;

[0088] Figure 6 It is a flow chart of the spiral signal inversion algorithm in the present invention;

[0089] Figure 7 This is the flow chart of the conversion between clock drive and mileage drive in the spiral signal inversion algorithm;

[0090] Figure 8 It is a schematic diagram of the pipeline column coordinate system model and the side expansion of the pipeline scanning trajectory;

[0091] Fig. 9 This is an expanded diagram of a pipeline inner wall detection path based on a spiral signal inversion algorithm of the present invention;

[0092] Fig.10 The present invention provides a pipeline plane contour map and a pipeline three-dimensional defect map after defect data correction. DETAILED DESCRIPTION

[0093] The present invention will be further described below in conjunction with the accompanying drawings and embodiments;

[0094] The invention designs a pipeline detection robot based on spiral scanning and a detection method thereof.

[0095] A pipeline inspection robot based on spiral scanning, as shown in the attached Figure 1-4 As shown; it includes a crawler body 1, a spiral detection system 2, a sensing mechanism 4 and a control mechanism 3; the spiral detection system is installed on the rear half of the body, the sensing mechanism is connected to the control mechanism, the control mechanism is connected to the spiral detection system, and the sensing mechanism and the control mechanism are both connected to the crawler body;

[0096] Tracked vehicle body 1, as attached Figure 1 , Attachment Figure 2As shown, it includes a vehicle body frame 1-1, a driving wheel 1-2, a driven wheel 1-3, a driving motor 1-4, a driving motor fixing frame 1-5, a bevel gear set 1-6, a ball screw 1-7, a multi-connection screw nut 1-8, a first screw support plate 1-9, a second screw support plate 1-10, an active diameter-changing rod 1-11, a mileage wheel fixing groove 1-12, a first driven diameter-changing rod and a second driven diameter-changing rod 1-13, a first track fixing plate and a second track fixing plate 1-14, a track inner baffle plate and a track outer baffle plate 1-15, a bearing 1-16, and an excitation coil fixing plate 1-17 , two driving wheels 1-2 and two driven wheels 1-3 are divided into two groups and symmetrically arranged on both sides of the vehicle body. The two driving wheels 1-2 are respectively connected to the main gears of the two bevel gear sets 1-6, and the sub gears of the two bevel gear sets 1-6 are respectively connected to the two driving motors 1-4. The bevel gear sets 1-6 are used to change the direction of motor rotation and improve the transmission ratio. The driving motor 1-4 is fixed to the inner baffle plate 1-15 of the track through the driving motor fixing frame 1-5. The driving wheel 1-2 and the driven wheel 1-3 are connected to the inner baffle plate and the outer baffle plate 1-15 of the track through the bearing 1-16. The baffle plate and the track outer baffle plate 1-15 are respectively connected to the first driven diameter reducing rod and the second driven diameter reducing rod 1-13 through the first track fixing plate and the second track fixing plate 1-14, and are connected to the vehicle body frame 1-1 through the first driven diameter reducing rod and the second driven diameter reducing rod 1-13. The two ends of the ball screw 1-7 are respectively fixed to the first screw support plate 1-9 and the second screw support plate 1-10. The multi-connection screw nut 1-8 is sleeved on the ball screw 1-7 and can move forward and backward along it. The three active diameter reducing rods 1-11 are connected to the side of the multi-connection screw nut 1-8. The two lower The active reducing rod 1-11 is connected to the second track fixing plate 1-14. When the multi-connection lead screw nut 1-8 moves along the ball screw, the active reducing rod 1-11 can be driven to change its angle, thereby pulling the first driven reducing rod and the second driven reducing rod 1-13 to achieve the purpose of expanding or contracting the track to adapt to different pipe diameters. A mileage wheel fixing groove is installed on the top of the active reducing rod 1-11 located above, which is convenient for installing the mileage wheel structure for calculating distance and positioning. The excitation coil fixing plate 1-17 is fixed to the front end of the first lead screw support plate 1-9 for installing the far-field eddy current excitation coil.

[0097] An active reducing rod is installed above the ball screw, which also expands or contracts with the change of the ball screw. A mileage wheel fixing groove is installed on the top of the active reducing rod to facilitate the installation of the mileage wheel structure for distance calculation and positioning.

[0098] The excitation coil fixing plate is fixed to the front end of the first lead screw support plate and is used for installing the remote field eddy current excitation coil;

[0099] The spiral detection system 2 includes a detection arm lifting platform and a cross-type detection mechanism. The detection arm lifting platform, as shown in the attached Figure 3 As shown, it includes a lifting platform base 2-1, a steering gear 2-2, a steering gear gear 2-3, a telescopic ruler bar 2-4, and a rotating motor bracket 2-5. The cross-type detection mechanism, as shown in the attached Figure 4 As shown, it includes a rotating motor 2-6, a coupling 2-7, a central connecting ring 2-8, n Detection arms, each of which includes a sleeve 2-9, a telescopic rod 2-10, an elastic structure 2-11, and a detection probe 2-12. The lifting platform base 2-1 is fixed on the upper surface of the vehicle frame 1-1. The telescopic ruler bar 2-4 is connected to the lifting platform base and can slide relatively. The steering gear 2-2 contacts the telescopic ruler bar 2-4 through the steering gear gear 2-3. The rotation angle of the steering gear 2-2 is adjusted. The rotation of the steering gear gear 2-3 can drive the telescopic ruler bar 2-4 to rise or fall to a precise position, thereby realizing the control of the height of the cross-type detection mechanism. The rotating motor 2-6 is fixed to the top of the telescopic ruler bar 2-4 through the rotating motor bracket 2-5 and can move up and down with it. The central connecting ring 2-8 of the cross-type detection mechanism is connected to the rotating motor 2-6 through the coupling 2-7. n The detection arms are distributed at an angle of 90 degrees outside the central connecting ring 2-8 and fixed thereto. The sleeve 2-9 of the detection arm is connected to the telescopic rod 2-10 through a slot structure. An elastic structure 2-11 is installed inside the sleeve to change the length of the detection arm. The other end of the telescopic rod 2-10 is fixed to the detection probe 2-12. The detection probe 2-12 reserves space for winding detection coils and wiring. When the rotating motor is working, the cross-type detection mechanism rotates, and the detection probe 2-12 moves close to the inner wall of the pipeline for scanning.

[0100] The control mechanism 3 includes an STM32F4 controller, a lithium battery power supply, a motor drive module, a DC step-down module, and a signal transmission module. The power supply is connected to the STM32F4 controller, the motor drive module, and the DC step-down module. The STM32F4 controller is connected to the motor drive module and the signal transmission module through wires, respectively. The DC step-down module and the motor drive module are both connected to the drive motor 1-4 and the rotating motor 2-6 through wires. The STM32F4 controller, the lithium battery power supply, the motor drive module, the DC step-down module, and the signal transmission module are installed on the body frame 1-1.

[0101] The sensing mechanism 4 includes an Mpu6050 electronic gyroscope, an infrared ranging module, and a Hall speed measurement module; the Mpu6050 electronic gyroscope is installed on the lower surface of the vehicle frame 1-1 and is parallel to the surface of the crawler vehicle body 1, the infrared ranging module is fixed on the rotating motor bracket 2-5 in the detection arm lifting platform, and the Hall speed measurement module is connected to the driving motor 1-4 and the rotating motor 2-6 of the crawler vehicle body.

[0102] A detection method of a pipeline detection robot based on spiral scanning is implemented based on the above pipeline detection robot based on spiral scanning, and comprises the following steps:

[0103] Step 1: Set the robot's drive motor initial speed n0, the ratio of the rotation motor speed to the drive motor speed K, the spiral detection system eccentricity threshold TH, and the sampling width h , detect the maximum allowable sampling width error e , where the ratio of the rotating motor speed to the driving motor speed K is a constant value determined by the maximum effective detection range of the detection probe, the eccentricity threshold TH of the spiral detection system is a constant value determined by the diameter range adapted to the small elastic diameter change of the detection arm, and the sampling width h It is a constant value determined by the sampling principle of the detection probe and detects the maximum allowable sampling width error. e It is a constant value determined by the scanning accuracy requirement of the detection probe;

[0104] Step 1.1: Establish a right-handed rectangular coordinate system inside the oil and gas pipeline to be inspected, where the robot movement direction along the central axis of the pipeline is the positive direction of the Z axis, as shown in the attached figure. Figure 8 , assuming that the cross-type detection mechanism rotates clockwise when viewed from the front of the robot; the movement of the center point of the detection area of ​​the detection probe on the inner wall of the pipeline is the combined motion of the linear motion along the Z axis and the circumferential circular motion in the OXY plane. For any initial position of the projection of the detection probe of a certain detection arm on the inner wall of the pipeline , its motion trajectory equation is:

[0105]

[0106] In the formula, is the rotational angular velocity of the detection arm in the cross-type detection mechanism, v is the robot's moving speed, t For working hours, x , y , z are respectively the intercepts of the projection position of the detection probe of the detection arm on the inner wall of the pipeline on the XYZ axis;

[0107] Step 1.2: Align the cylindrical surface where the pipe is located along a straight line x = R , y =0 and unfolded into a Cartesian coordinate system. The unfolded image of the inner surface of the pipe scanned by the robot is a rectangle with a length of L , width is W , as attached Figure 8 ;

[0108] and L = S ,W = C , C =2π R , S = vt ;

[0109] In the formula, S To detect the distance traveled by the vehicle during working hours, C is the circumference of the pipe section, R is the inner radius of the pipe;

[0110] The spiral detection trajectory will be unfolded into an inclined straight line in the Cartesian coordinate system, and the angle between the straight line and the length of the rectangle is for:

[0111]

[0112] The distance between the centers of two adjacent trajectories d for:

[0113]

[0114] In the formula, n is the number of detection arms involved in scanning in the spiral detection system. In this embodiment, n is 4;

[0115] The difference between the distance between the two detection tracks and the sampling width should be less than the maximum allowable sampling width error to avoid scanning omissions and scanning duplications. After calculation, the conditions that need to be met are:

[0116]

[0117] In the formula, e To detect the maximum allowable sampling width error, h is the sampling width;

[0118] Initially set the ratio K of the robot's rotation motor speed to the drive motor speed:

[0119]

[0120] In the formula, r is the radius of the robot's driving wheel;

[0121] In summary, in order to ensure that the detection arm fully scans the inner wall of the pipeline, in this embodiment, the ratio K of the rotating motor speed to the driving motor speed needs to be maintained to meet the following conditions through the control algorithm:

[0122]

[0123] Ideally, when detecting the maximum allowable sampling width error eWhen it is 0, the ratio K of the rotating motor speed to the driving motor speed can be calculated using the following formula:

[0124]

[0125] Step 2: The infrared ranging module of the sensing mechanism obtains the position information of the central connecting ring of the cross-type detection mechanism in the axial direction of the pipeline and feeds it back to the control mechanism. The control mechanism adjusts the rotation angle of the servo to control the height of the detection arm lifting platform. Through the axis center control method, the geometric center of the cross-type detection mechanism is always at the axial center of the pipeline. The algorithm flow is as shown in the attached figure. Figure 5 ;

[0126] Step 2.1: The infrared distance measuring module on the top of the detection arm lifting platform transmits the collected distance information from the inner wall of the pipeline to the control mechanism in real time;

[0127] Step 2.2: The control mechanism calculates the eccentricity X of the position of the center of the cross-type detection mechanism on the pipeline cross section according to the distance information obtained in real time;

[0128] Step 2.3: Compare the acquired relevant parameters with the set eccentricity threshold TH of the spiral detection system, control the lifting platform through the position PID algorithm, and adjust the rotation direction, angle and speed of the servo;

[0129] Step 2.4: Repeat steps 2.1 to 2.3 until the eccentricity X is always controlled to be less than or equal to the eccentricity threshold TH, and then perform output limiting;

[0130] Step 3: The control mechanism uses an incremental PID feedback algorithm to achieve speed closed-loop control according to the required setting value of the driving motor speed, and drives the driving motor of the crawler body and the rotating motor of the spiral detection system to work respectively. The Hall speed measurement module obtains the rotation speeds of the driving motor of the crawler body and the rotating motor of the spiral detection system, and transmits them to the control mechanism. The control mechanism compares the rotation speeds of the driving motor of the crawler body and the rotating motor of the spiral detection system obtained by the Hall speed measurement module with the initial rotation speed n0 of the driving motor of the robot and the ratio K of the rotating motor speed to the driving motor speed, and performs feedback control to achieve the set rotation speeds of the driving motor and the rotating motor, so as to ensure that the inner wall of the pipeline is scanned without omission during the travel process;

[0131] Step 4: During the robot's operation, as the robot moves and the cross-type detection mechanism rotates, the detection probe moves close to the inner wall of the pipeline to collect pipeline defect signals. However, since the signals collected by the cross-type detection mechanism are nThe detection probe has a spiral trajectory, which makes it impossible to directly analyze the defect voltage and location information of the defect. To solve the above problems, a spiral signal inversion algorithm is used to invert and correct the spiral signal, and convert the three-dimensional pipeline inner wall information into a two-dimensional pipeline inner wall expansion diagram to realize the visualization of pipeline inner wall defects.

[0132] In this embodiment, the test environment is selected as a metal pipe with a diameter of 219 mm, and the radius of the robot's driving wheel is about 24 mm. In order to ensure that the detection signal is not missed, the detection probe sampling interval is selected as 1 cm. Under this condition, the test pipe diameter and the detection probe sampling interval are substituted into step 1 to calculate the ratio of the rotating motor speed to the driving motor speed K, which is about 23.65, so that the rotating arm rotation angular velocity ( rad / s ) and the car's forward speed ( v / s ) is about 1.568. Then, according to the actual requirements, the rotation angular velocity is set to 10.0 rad / s, the forward speed is set to 6.4 cm / s, and in order to ensure high-precision sampling data, the sampling frequency is set to 573 Hz, ensuring that each rotating arm has 360 sampling points per rotation. After setting, enter the pipeline environment for detection.

[0133] Step 4.1: Use Matlab to read the sampling data of the detection probe and the vehicle body operation data from the data storage device, including n The voltage value measured by the detection arm, the phase of the detection arm, the time of the trolley movement, and the number of motor shaft rotation pulses measured by the Hall speed measurement module;

[0134] Step 4.2: Preliminarily identify the defect values ​​and abnormal values ​​of the sampled data, filter out the noise signals, and then perform data preprocessing such as fitting and interpolation on the preliminarily processed data; data with more decimal places should be approximated.

[0135] Step 4.3: Read the preprocessed data into the structure data type to form a data set, and use the new stacking method to process the data according to the actual physical meaning to increase the dimension of the data set, and finally add interactive features to each sampled pipe wall signal, including the phase when measuring the value, time data, and the number of motor shaft rotation pulses;

[0136] Step 4.4: The clock drive is converted into a mileage drive algorithm to obtain the mileage of the car. At the same time, the electronic gyroscope signal on the car body is used to correct the signal of the detection arm to reduce the detection data error caused by the possible tilt of the car body. After correction, it is inserted into the data set to finally obtain a database of sampled data;

[0137] The flowchart of the algorithm for converting clock drive into mileage drive in the spiral signal inversion algorithm is shown in the attached figure. Figure 7 , the specific steps are:

[0138] Step 4.4.1: Extract the data set of the vehicle movement time and the motor shaft rotation pulse number;

[0139] Step 4.4.2: Read the number of encoder pulses per unit clock measured by the Hall speed measurement module and calculate the forward speed of the car;

[0140] Step 4.4.3: Calculate the mileage of the car by using the step-by-step integration algorithm from the car's forward speed;

[0141] Step 4.4.4: Compare and correct the mileage of the vehicle obtained from steps 4.4.1 to 4.4.3 with the mileage data collected by the sensing mechanism to reduce the accidental errors caused by these two methods;

[0142] Step 4.5: Use the loop structure to use the vehicle body inclination information collected by the electronic gyroscope as the characteristic value index, and find the voltage values ​​of different spiral arms at different mileage coordinates of the phase in the database in turn, and finally obtain the database of each phase value. After drawing, the pipeline inner wall expansion diagram is obtained, as shown in the attached figure. Fig. 9 As shown in the figure, the data visualization work is completed; at the same time, the defect contour map can be drawn to more intuitively see the defect size, depth and other characteristic information, and the defect three-dimensional map can be drawn, as shown in the attached figure. Fig.10 shown.

[0143] Step 5: In order to further judge the type of defects and the specific location of the defects in the pipeline inner wall expansion diagram, a pipeline inner wall defect recognition model based on deep learning is proposed. In the output results of the model, the defect feature information judgment algorithm is used to identify and mark the geometric features of the defects, including the defect area, perimeter, maximum boundary distance, and maximum depth. At the same time, the distance and phase information of the defect in the pipeline are marked, and finally the pipeline inner wall expansion diagram after processing is output;

[0144] Step 5.1: Obtain data by collecting the pipeline inner wall development diagram obtained in step 4;

[0145] Step 5.2: First, perform preliminary cropping on the original image, and then perform noise filtering and image enhancement preprocessing on the cropped image;

[0146] The cropping is to use a screenshot tool to crop the original image to obtain the target measurement area, and the image is saved in "jpg" format;

[0147] Step 5.3: Train the target detection network model (Faster-RCNN) based on the region proposal algorithm and obtain the pipeline inner wall defect recognition model;

[0148] Step 5.3.1: Initialize each module of the neural network’s unique layer parameters using pre-trained weights from a large-scale image recognition database (ImageNet);

[0149] Step 5.3.2: Collect inspection images of normal pipeline inner wall and pipeline inner wall with defects, classify and mark them, determine training set and test set, and form a sample library;

[0150] Step 5.3.2.1: Use the detection probe to detect various known types of defects and the inner wall of the normal pipeline, and generate a detection image;

[0151] Step 5.3.2.2: Identify and label the defects on the inner wall of the pipeline; select the defect images with obvious features, remove the invalid data that are difficult to identify or have been repaired, and segment the selected images using a fixed pixel size. Then use the data labeling tool software Labelme to label the types of all defects. The labels correspond to the defect types one by one to obtain a set of labeling sets;

[0152] Step 5.3.2.3: Determine the training set and test set: According to the number of samples, divide the training samples into training set and test set in a ratio of 7:3, integrate the txt files, pictures and xml files into files that can be read by Faster_RCNN to form a sample library;

[0153] Step 5.3.3: The training is divided into two parts, one is to train the RPN network, and the other is to train the subsequent classification network; use the data set in step 5.3.2 to train the parameters of each network layer in turn;

[0154] Step 5.3.4: After testing the test set, fine-tune the parameters of each network layer according to the test results and the actual error, and finally form a unified network to obtain the pipeline inner wall defect recognition model;

[0155] Step 5.4: Use the model to perform defect identification process, determine whether there are defects on the inner wall of the pipeline, frame the defect location and mark the type;

[0156] Step 5.4.1: Scale the detection image to a fixed size and then pass it into the feature extraction network. The feature extraction network uses a deep residual network to obtain a feature map;

[0157] Step 5.4.2: The obtained feature map is passed to the RPN part to generate a proposed candidate box where defects may occur, determine whether there are defects in the candidate box, and make the range of the candidate box more accurate by correcting the box;

[0158] Step 5.4.3: Extract the defect suggestion candidate box generated in step 5.4.2 and the defect feature map in step 5.4.1, and pass them to the ROI Polling layer to obtain a fixed-size feature map containing the candidate box and send it to the subsequent fully connected layer for classification and regression;

[0159] Step 5.4.4: Send the feature map of the fixed-size candidate box obtained in step 5.4.3 to the subsequent classification layer and regression layer for classification and regression operations to obtain the specific location of the target;

[0160] Step 5.5: Determine the defect feature information including geometric feature information and orientation feature information based on the defect feature information processing algorithm;

[0161] Step 5.5.1: Get the framed defects identified by Faster R-CNN;

[0162] Step 5.5.2: By means of image enhancement, the image is converted into a grayscale image through grayscale transformation;

[0163] Step 5.5.3: Use the median filter method to perform noise filtering on the grayscale image, and then perform morphological processing on the grayscale image after noise filtering, including corrosion and expansion operations;

[0164] Step 5.5.4: Use the Canny edge detection algorithm to determine the defect edge position;

[0165] Step 5.5.5: Determine the target defect in the image, and then obtain the total number of pixels of the target defect. By comparing it with the number of pixels of the standard object, various geometric feature information of the defect is finally obtained, including the defect area, perimeter, and maximum boundary distance.

[0166] Step 5.5.6: Recall the defect-related information in the database, and after conversion, obtain the maximum depth or maximum bulge height of the defect, as well as the azimuthal characteristic information of the defect in the pipeline, including the position and phase of the defect in the pipeline, and finally add all the characteristic information to the annotation.

[0167] Step 5.6: Output the pipeline inner wall defect expansion diagram after processing.

Claims

1. A pipeline inspection robot based on spiral scanning, characterized in that: It includes a crawler-type vehicle body, a spiral detection system, a sensing mechanism and a control mechanism; the spiral detection system is installed on the rear half of the vehicle body, the sensing mechanism is connected to the control mechanism, the control mechanism is connected to the spiral detection system, and the sensing mechanism and the control mechanism are both connected to the crawler-type vehicle body; The spiral detection system includes a detection arm lifting platform and a cross-type detection mechanism. The detection arm lifting platform includes a lifting platform base, a steering gear, a steering gear gear, a telescopic ruler bar, and a rotating motor bracket; the cross-type detection mechanism includes a rotating motor, a coupling, a center connecting ring, n detection arms, each of which includes a sleeve, a telescopic rod, an elastic structure, and a detection probe. The lifting platform base is fixed on the upper surface of the vehicle frame. The telescopic ruler bar is connected to the lifting platform base and slides relatively. The steering gear contacts the telescopic ruler bar through the steering gear gear. The rotating motor is fixed to the top of the telescopic ruler bar through the rotating motor bracket and moves up and down with it. The central connecting ring of the cross-type detection mechanism is connected to the rotating motor through a coupling. n The detection arms are Angle Distributed outside the central connecting ring and fixed thereto, the sleeve of the detection arm is connected to the telescopic rod through a slot structure, an elastic structure is installed inside the sleeve to change the length of the detection arm, the other end of the telescopic rod is fixed to the detection probe, and the detection probe reserves space for winding the detection coil and wiring; The control mechanism includes a controller, a power supply, a motor drive module, a DC step-down module, and a signal transmission module. The power supply is connected to the controller, the motor drive module, and the DC step-down module. The controller is connected to the motor drive module and the signal transmission module through wires, respectively. The DC step-down module and the motor drive module are both connected to the drive motor and the rotating motor through wires. The controller, the power supply, the motor drive module, the DC step-down module, and the signal transmission module are installed on the vehicle body frame. The sensing mechanism includes an electronic gyroscope, an infrared ranging module, and a Hall speed measurement module; the electronic gyroscope is installed on the lower surface of the vehicle frame and is parallel to the surface of the crawler vehicle body, the infrared ranging module is fixed on the rotating motor bracket in the detection arm lifting platform, and the Hall speed measurement module is connected to the driving motor and the rotating motor of the crawler vehicle body; The spiral scanning pipeline inspection robot realizes pipeline inspection based on the following methods: Step 1: Set the robot's drive motor initial speed n0, the ratio of the rotation motor speed to the drive motor speed K, the spiral detection system eccentricity threshold TH, and the sampling width h , detect the maximum allowable sampling width error e , where the ratio of the rotating motor speed to the driving motor speed K is a constant value determined by the maximum effective detection range of the detection probe, the eccentricity threshold TH of the spiral detection system is a constant value determined by the diameter range adapted to the small elastic diameter change of the detection arm, and the sampling width h It is a constant value determined by the sampling principle of the detection probe and detects the maximum allowable sampling width error. e It is a constant value determined by the scanning accuracy requirement of the detection probe; Step 2: The infrared ranging module of the sensing mechanism obtains the position information of the central connecting ring of the cross-type detection mechanism in the axial direction of the pipeline and feeds it back to the control mechanism. The control mechanism adjusts the rotation angle of the servo to control the height of the detection arm lifting platform. Through the axis center control method, the geometric center of the cross-type detection mechanism is always at the axial center of the pipeline. Step 3: The control mechanism uses an incremental PID feedback algorithm to achieve speed closed-loop control according to the required setting value of the driving motor speed, and drives the driving motor of the crawler body and the rotating motor of the spiral detection system to work respectively. The Hall speed measurement module obtains the rotation speeds of the driving motor of the crawler body and the rotating motor of the spiral detection system, and transmits them to the control mechanism. The control mechanism compares the rotation speeds of the driving motor of the crawler body and the rotating motor of the spiral detection system obtained by the Hall speed measurement module with the initial rotation speed n0 of the driving motor of the robot and the ratio K of the rotating motor speed to the driving motor speed, and performs feedback control to achieve the set rotation speeds of the driving motor and the rotating motor, so as to ensure that the inner wall of the pipeline is scanned without omission during the travel process; Step 4: When the robot is working, the detection probe moves in a spiral trajectory close to the inner wall of the pipeline to collect signals. The spiral signal is inverted and corrected through a set of spiral signal inversion algorithms, and the three-dimensional pipeline inner wall information is converted into a two-dimensional pipeline inner wall expansion diagram to realize the visualization of pipeline inner wall defects; Step 5: In order to further judge the type of defects and the specific location of the defects in the pipeline inner wall expansion diagram, a pipeline inner wall defect recognition model based on deep learning is proposed. In the output results of the model, the defect feature information judgment algorithm is used to identify and mark the geometric features of the defects, including the defect area, perimeter, maximum boundary distance, and maximum depth. At the same time, the distance and phase information of the defect in the pipeline are marked, and finally the pipeline inner wall expansion diagram after processing is output.

2. The pipeline inspection robot based on spiral scanning according to claim 1, characterized in that: The crawler vehicle body comprises a vehicle body frame, a driving wheel, a driven wheel, a driving motor, a driving motor fixing frame, a bevel gear set, a ball screw, a multi-connection screw nut, a first screw support plate, a second screw support plate, a driving reducer, a mileage wheel fixing groove, a first driven reducer and a second driven reducer, a first track fixing plate and a second track fixing plate, a track inner baffle plate and a track outer baffle plate, a bearing, and an excitation coil fixing plate; the two driving wheels and the two driven wheels are divided into two groups and symmetrically arranged on both sides of the vehicle body, the two driving wheels are respectively connected to the main gears of the two bevel gear sets, the sub gears of the two bevel gear sets are respectively connected to the two driving motors, and the driving motors The driving motor fixing frame is fixed to the inner fender of the track, the driving wheel and the driven wheel are connected to the inner fender of the track and the outer fender of the track through bearings, the inner fender of the track and the outer fender of the track are respectively connected to the first driven reducing rod and the second driven reducing rod through the first track fixing plate and the second track fixing plate, and are connected to the vehicle body frame through the first driven reducing rod and the second driven reducing rod, the two ends of the ball screw are respectively fixed to the first screw support plate and the second screw support plate, the multi-connection screw nut is sleeved on the ball screw and moves forward and backward along the ball screw, the three active reducing rods are connected to the side of the multi-connection screw nut, and the two active reducing rods located at the bottom are connected to the second track fixing plate; An active reducing rod is installed above the ball screw, a mileage wheel fixing groove is installed on the top of the active reducing rod, and an exciting coil fixing plate is fixed on the front end of the first screw support plate.

3. The pipeline inspection robot based on spiral scanning according to claim 1, characterized in that: Step 1 is as follows: Step 1.1: Establish a right-handed rectangular coordinate system inside the oil and gas pipeline to be inspected, where the robot movement direction along the central axis of the pipeline is the positive direction of the Z axis. Assuming that the cross-type detection mechanism rotates clockwise when viewed from the front of the robot, the movement of the center point of the detection area of ​​the detection probe on the inner wall of the pipeline is the combined motion of the linear motion along the Z axis and the circumferential circular motion in the OXY plane. For any initial position of the projection of the detection probe of a certain detection arm on the inner wall of the pipeline , its motion trajectory equation is: ； In the formula, is the rotational angular velocity of the detection arm in the cross-type detection mechanism, v is the robot's moving speed, t For working hours, x , y , z are respectively the intercepts of the projection position of the detection probe of the detection arm on the inner wall of the pipeline on the XYZ axis; Step 1.2: Align the cylindrical surface where the pipe is located along a straight line x = R , y =0 and unfolded into a Cartesian coordinate system. The unfolded image of the inner surface of the pipe scanned by the robot is a rectangle with a length of L , width is W ; and L = S , W = C , C =2π R , S = vt ; In the formula, S To detect the distance traveled by the vehicle during working hours, C is the circumference of the pipe section, R is the inner radius of the pipe; The spiral detection trajectory will be unfolded into an inclined straight line in the Cartesian coordinate system, and the angle between the straight line and the length of the rectangle is for: ； The distance between the centers of two adjacent trajectories d for: ； In the formula, n is the number of detection arms involved in scanning in the spiral detection system; The difference between the distance between the two detection tracks and the sampling width should be less than the maximum allowable sampling width error to avoid scanning omissions and scanning duplications. After calculation, the conditions that need to be met are: ； In the formula, e To detect the maximum allowable sampling width error, h is the sampling width; Initially set the ratio K of the robot's rotation motor speed to the drive motor speed: ； In the formula, r is the radius of the robot's driving wheel; In summary, in order to ensure that the detection arm fully scans the inner wall of the pipeline, the control algorithm needs to be used to maintain the ratio K of the rotating motor speed to the driving motor speed to meet the following conditions: ； Ideally, when detecting the maximum allowable sampling width error e When it is 0, the ratio K of the rotating motor speed to the driving motor speed is calculated using the following formula: 。 4. The pipeline inspection robot based on spiral scanning according to claim 1, characterized in that: Step 2 is as follows: Step 2.1: The infrared distance measuring module on the top of the detection arm lifting platform transmits the collected distance information from the inner wall of the pipeline to the control mechanism in real time; Step 2.2: The control mechanism calculates the eccentricity X of the position of the center of the cross-type detection mechanism on the pipeline cross section according to the distance information obtained in real time; Step 2.3: Compare the acquired relevant parameters with the set eccentricity threshold TH of the spiral detection system, control the lifting platform through the position PID algorithm, and adjust the rotation direction, angle and speed of the servo; Step 2.4: Repeat steps 2.1 to 2.3 until the eccentricity X is controlled to be always less than or equal to the eccentricity threshold TH, and then perform output limiting.

5. The pipeline inspection robot based on spiral scanning according to claim 1, characterized in that: Step 4 is as follows: Step 4.1: Use Matlab to read the sampling data of the detection probe and the vehicle body operation data from the data storage device, including n The voltage value measured by the detection arm, the phase of the detection arm, the time of the trolley movement, and the number of motor shaft rotation pulses measured by the Hall speed measurement module; Step 4.2: Preliminarily identify the defect values ​​and abnormal values ​​of the sampled data, filter out the noise signal, and then fit the preliminarily processed data, perform interpolation data preprocessing, and perform approximate processing on the data with more decimal places; Step 4.3: Read the preprocessed data into the structure data type to form a data set, and use the new stacking method to process the data according to the actual physical meaning to increase the dimension of the data set, and finally add interactive features to each sampled pipe wall signal, including the phase when measuring the value, time data, and the number of motor shaft rotation pulses; Step 4.4: The clock drive is converted into a mileage drive algorithm to obtain the mileage of the car. At the same time, the electronic gyroscope signal on the car body is used to correct the signal of the detection arm to reduce the detection data error caused by the tilt of the car body. After correction, it is inserted into the data set to finally obtain a database of sampled data; The specific steps of converting the clock drive into the mileage drive algorithm in the spiral signal inversion algorithm are as follows: Step 4.4.1: Extract the data set of the vehicle movement time and the motor shaft rotation pulse number; Step 4.4.2: Read the number of encoder pulses per unit clock measured by the Hall speed measurement module and calculate the forward speed of the car; Step 4.4.3: Calculate the mileage of the car by using the step-by-step integration algorithm from the car's forward speed; Step 4.4.4: Compare and correct the mileage of the car obtained from steps 4.4.1 to 4.4.3 with the mileage data collected by the sensing mechanism to reduce the accidental errors caused by these two methods; Step 4.5: Use the loop structure to use the vehicle body inclination information collected by the electronic gyroscope as the characteristic value index, and find the voltage values ​​of different detection arms at different mileage coordinates of the phase in the database in turn, and finally obtain the database of each phase value. After drawing, the pipeline inner wall expansion diagram is obtained to complete the data visualization work.

6. The pipeline inspection robot based on spiral scanning according to claim 1, characterized in that: Step 5 is as follows: Step 5.1: Obtain data by collecting the pipeline inner wall development diagram obtained in step 4; Step 5.2: First, perform preliminary cropping on the original image, and then perform noise filtering and image enhancement preprocessing on the cropped image; The cropping is to use a screenshot tool to crop the original image to obtain the target measurement area, and the image is saved in "jpg" format; Step 5.3: Train the target detection network model (Faster-RCNN) based on the region proposal algorithm and obtain the pipeline inner wall defect recognition model; Step 5.4: Use the model to perform defect identification process, determine whether there are defects on the inner wall of the pipeline, frame the defect location and mark the type; Step 5.5: Determine the defect feature information including geometric feature information and orientation feature information based on the defect feature information processing algorithm; Step 5.5.1: Get the framed defects identified by Faster R-CNN; Step 5.5.2: By means of image enhancement, the image is converted into a grayscale image through grayscale transformation; Step 5.5.3: Use the median filter method to perform noise filtering on the grayscale image, and then perform morphological processing on the grayscale image after noise filtering, including corrosion and expansion operations; Step 5.5.4: Use the Canny edge detection algorithm to determine the defect edge position; Step 5.5.5: Determine the target defect in the image, and then obtain the total number of pixels of the target defect. By comparing it with the number of pixels of the standard object, various geometric feature information of the defect is finally obtained, including the defect area, perimeter, and maximum boundary distance; Step 5.5.6: Recall the defect information in the database, convert it, and obtain the maximum depth or maximum bulge height of the defect, as well as the position characteristic information of the defect in the pipeline, including the position and phase of the defect in the pipeline, and finally add all the characteristic information to the annotation; Step 5.6: Output the pipeline inner wall defect expansion diagram after processing.

7. The pipeline inspection robot based on spiral scanning according to claim 6, characterized in that: Step 5.3 is as follows: Step 5.3.1: Initialize each module of the neural network's unique layer parameters using pre-trained weights from a large-scale image recognition database (ImageNet); Step 5.3.2: Collect inspection images of normal pipeline inner wall and pipeline inner wall with defects, classify and mark them, determine training set and test set, and form a sample library; Step 5.3.2.1: Use the detection probe to detect various known types of defects and the inner wall of the normal pipeline, and generate a detection image; Step 5.3.2.2: Identify and label the defects on the inner wall of the pipeline; select the defect images with obvious features, remove the invalid data that are difficult to identify or have been repaired, and segment the selected images using a fixed pixel size. Then use the data labeling tool software Labelme to label the types of all defects. The labels correspond to the defect types one by one to obtain a set of labeling sets; Step 5.3.2.3: Determine the training set and test set: According to the number of samples, divide the training samples into training set and test set in a ratio of 7:3, integrate the txt files, pictures and xml files into files that can be read by Faster_RCNN to form a sample library; Step 5.3.3: The training is divided into two parts, one is to train the RPN network, and the other is to train the subsequent classification network; use the data set in step 5.3.2 to train the parameters of each network layer in turn; Step 5.3.4: After testing the test set and fine-tuning the parameters of each network layer based on the test results and the actual error, a unified network is finally formed to obtain the pipeline inner wall defect recognition model.

8. The pipeline inspection robot based on spiral scanning according to claim 6, characterized in that: Step 5.4 is as follows: Step 5.4.1: Scale the detection image to a fixed size and then pass it into the feature extraction network. The feature extraction network uses a deep residual network to obtain a feature map; Step 5.4.2: The obtained feature map is passed to the RPN part to generate a proposed candidate box where defects may occur, determine whether there are defects in the candidate box, and make the range of the candidate box more accurate by correcting the box; Step 5.4.3: Extract the defect suggestion candidate box generated in step 5.4.2 and the defect feature map in step 5.4.1, and pass them to the ROI Polling layer to obtain a fixed-size feature map containing the candidate box and send it to the subsequent fully connected layer for classification and regression; Step 5.4.4: Send the feature map of the fixed-size candidate box obtained in step 5.4.3 to the subsequent classification layer and regression layer for classification and regression operations to obtain the specific location of the target.

Images