Tunnel segment joint quality high-precision detection method and system based on image-stabilizing mechanical arm

Abstract

The application discloses a tunnel segment joint quality high-precision detection method and system based on a steady image mechanical arm, and the device comprises a ring rail cruising platform, a self-adaptive steady image mechanical arm and a macro-micro fusion detection module; the ring rail cruising platform comprises a cruising platform pedestal and a guide rail, the guide rail is mounted on a shield machine tail H-shaped beam through a cushioning connecting piece, the guide rail is in a circular ring structure and located in the shield machine, and the cruising platform pedestal is arranged on the guide rail and can move along the circumference of the guide rail; one end of the self-adaptive steady image mechanical arm is mounted on the cruising platform pedestal, and the macro-micro fusion detection module is mounted on the other end of the self-adaptive steady image mechanical arm, so that the macro-micro fusion detection module moves synchronously with the cruising platform pedestal and is arranged to face the joint between a to-be-installed segment and an installed segment. The application relates to the technical field of tunnel construction, and can solve the problems of low efficiency, high danger, lag, large interference, low precision and poor flexibility of fixed equipment detection in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of tunnel construction, and in particular to a tunnel segment joint quality high-precision detection method and system based on a steady image mechanical arm. BACKGROUND

[0002] In modern shield tunnel construction, segment assembly quality is the "lifeline" of the tunnel structure. Among them, segment misalignment (radial and circumferential height difference between adjacent segments) is the highest quality problem. Small misalignment not only damages the waterproof seal and causes tunnel leakage, but also causes stress concentration, which poses a major safety hazard for long-term segment cracking and structural failure. Therefore, "real-time and accurate" detection of misalignment during assembly is a rigid requirement for ensuring project quality.

[0003] The existing segment misalignment detection methods mainly include manual detection and fixed equipment detection. The manual detection has the following technical problems:

[0004] High-risk operation: workers must enter the shield tail in a complex environment with dust, noise, and large machinery (such as segment transport vehicles), facing high collision and extrusion safety risks.

[0005] Low efficiency: relying on the human eye, touch, and measuring tools, the labor intensity is high, the speed is slow, and the detection data is highly subjective and has large errors.

[0006] Detection lag: often after the entire ring is assembled, the problem is found, and the cost of adjustment and rework is extremely high.

[0007] The fixed equipment detection has the following technical problems:

[0008] Large vibration interference: the detection equipment (such as a laser scanner) installed with a fixed support is rigidly connected with the shield machine. The propulsion, assembly, and slag removal of the shield machine will produce continuous strong vibration, causing the detection equipment to be unstable and the data to jump, and the measurement accuracy cannot be guaranteed.

[0009] Poor flexibility: unable to work with the shield machine, difficult to adjust the detection position, difficult to adapt to different diameter tunnels, and has a large number of detection blind areas.

[0010] Low integration: a separate support needs to be built, which occupies valuable operating space in the tunnel.

[0011] Therefore, it is necessary to provide a tunnel segment joint quality high-precision detection method and system based on a steady image mechanical arm, which can solve the problems of low efficiency, high risk, and lag in manual detection, and large interference, low precision, and poor flexibility in fixed equipment detection. SUMMARY

[0012] The application aims to provide a tunnel segment joint quality high-precision detection method and system based on a steady image mechanical arm, which can solve the problems of low efficiency, high risk, lag, large interference, low precision and poor flexibility of fixed equipment detection in the prior art.

[0013] The application is implemented as follows:

[0014] A tunnel segment joint quality high-precision detection system based on a steady image mechanical arm, comprising a ring rail cruising platform, an adaptive steady image mechanical arm and a macro-micro fusion detection module; the ring rail cruising platform comprises a cruising platform pedestal and a guide rail, the guide rail is installed on an H-shaped beam at the tail of a shield machine through a shock absorption connecting piece, the guide rail has a circular ring structure and is located inside the shield machine, and the cruising platform pedestal is arranged on the guide rail and can move along the circumference of the guide rail; one end of the adaptive steady image mechanical arm is installed on the cruising platform pedestal, and the macro-micro fusion detection module is installed on the other end of the adaptive steady image mechanical arm, so that the macro-micro fusion detection module can move synchronously with the cruising platform pedestal and face the joint between a to-be-installed segment and an installed segment.

[0015] The guide rail comprises rail-holding guide rollers, a ring rail base, a guide slide rail, a cruising driving system and a precision toothed rail; the guide slide rail and the precision toothed rail both have a circular ring structure and are concentrically stacked on the ring rail base and fixed, and the ring rail base is installed on the H-shaped beam at the tail of the shield machine through a shock absorption connecting piece; a pair of rail-holding guide rollers are arranged below the cruising platform pedestal in a spaced manner, and the rail-holding guide rollers are matched with the inner ring and the outer ring of the guide slide rail to roll and engage; the cruising driving system is installed on the cruising platform pedestal, the cruising driving system is in meshing transmission connection with the precision toothed rail; and the adaptive steady image mechanical arm is installed on the cruising platform pedestal.

[0016] The adaptive steady image mechanical arm comprises an end effector interface clamp, a mechanical arm body and an adaptive edge computing module; the base of the mechanical arm body is installed on the cruising platform pedestal of the ring rail cruising platform, the mechanical arm body is composed of a plurality of arm joints driven by servo motors in series, and an inertial measurement sensor is embedded in each joint; the adaptive edge computing module is arranged on and connected with the mechanical arm body, the adaptive edge computing module is integrated with a segment misalignment detection algorithm and an active dynamic motion decoupling and disturbance suppression algorithm; the end effector interface clamp is installed at the end of the mechanical arm body away from the cruising platform pedestal, the end effector interface clamp is integrated with a pitch joint and a rotation joint, and the macro-micro fusion detection module is installed at the end of the mechanical arm body through the pitch joint and the rotation joint.

[0017] The macro-micro fusion detection module comprises a secondary active stabilizing holder, a global array camera and a three-dimensional infrared scanner; the secondary active stabilizing holder is detachably connected with the end effector interface clamp jaw, and the global array camera and the three-dimensional infrared scanner are carried on the secondary active stabilizing holder, and the global array camera and the three-dimensional infrared scanner are arranged to face the joint between the to-be-installed segment and the installed segment.

[0018] A detection method of a tunnel segment joint quality high-precision detection system based on a stabilizing mechanical arm, comprising the following steps:

[0019] Step 1: The to-be-installed segment is positioned, that is, the to-be-installed segment is transported to the assembling position;

[0020] Step 2: The detection system receives a detection instruction, and the ring rail cruising platform drives the adaptive stabilizing mechanical arm to move to monitor the positioning process of the to-be-installed segment;

[0021] Step 3: The to-be-installed segment is assembled by the segment assembling machine;

[0022] Step 4: The ring rail cruising platform drives the adaptive stabilizing mechanical arm to move along the guide rail, and the adaptive stabilizing mechanical arm controls the macro-micro fusion detection module at the end thereof to move to a target detection area;

[0023] Step 5: The macro-micro fusion detection module performs preliminary detection on the target detection area at the joint of the to-be-installed segment;

[0024] Step 6: Based on the preliminary detection result, accurate positioning and stabilizing locking are performed by the adaptive stabilizing mechanical arm and the macro-micro fusion detection module;

[0025] Step 7: The macro-micro fusion detection module performs microscopic measurement and data acquisition on the target detection area at the joint;

[0026] Step 8: The to-be-installed segments of the whole ring are detected in sequence according to the segment assembling sequence, after the joint detection of one ring segment is completed, steps 1-7 are repeated, and a quality detection report of each joint is generated in real time;

[0027] Step 9: After the detection of all to-be-installed segments is completed, the adaptive stabilizing mechanical arm is contracted to a transportation posture, and the cruising platform base returns to the initial position along the guide rail.

[0028] The step 2 comprises the following sub-steps:

[0029] Step 21: When the to-be-installed segment is transported to the assembling position, the ring rail cruising platform drives the adaptive stabilizing mechanical arm and the macro-micro fusion detection module to move to a predetermined observation position under the control of the adaptive edge computing module;

[0030] Step 22: The adaptive image stabilization robotic arm maintains stable tracking during the movement of the segment to be installed through an active dynamic motion decoupling and disturbance suppression algorithm;

[0031] Step 23: The global array camera of the macro-micro fusion detection module continuously captures images of the moving tube segment to be installed under the image stabilization guarantee of the secondary active gimbal, and transmits the real-time image data stream to the adaptive edge computing module to realize the monitoring of the placement process of the tube segment to be installed.

[0032] In step 5, the macro-micro fusion detection module performs a rapid macro-scan of the circumferential and tangential seams of the segment to be installed: the global array camera acquires panoramic images and transmits the panoramic image data to the adaptive edge computing module in real time for preliminary seam identification, and records the macro position and orientation data of the segment to be installed.

[0033] In step 6, based on the preliminary detection results in step 5, namely the macroscopic position and attitude data of the segment to be installed, the fine detection stage begins: the adaptive edge computing module calculates the precise three-dimensional coordinates of the joint through the hand-eye calibration matrix, the ring-track patrol platform performs coarse positioning, the adaptive image-stabilized robotic arm is moved to the target area, the robotic arm body of the adaptive image-stabilized robotic arm performs fine pose adjustment through the movement of each arm joint, and performs micro-angle compensation through the pitch and rotation joints built into the end effector interface gripper, and the secondary active stabilization gimbal activates high-frequency vibration suppression to achieve pixel-level stabilization.

[0034] In step 7, the adaptive edge computing module sends a measurement command to the macro-micro fusion detection module, the three-dimensional infrared scanner performs high-density point cloud acquisition on the target joint, and transmits all the original three-dimensional point cloud data to the adaptive edge computing module without loss. The segment misalignment detection algorithm processes the data in real time and calculates the misalignment amount and joint width.

[0035] In step 8, after the joint inspection of one ring is completed, the entire inspection system moves with the tunnel boring machine to the assembly position of the next ring segment and automatically plans the joint inspection path of the next ring segment. The patrol platform base moves precisely to the next inspection position along the guide rail.

[0036] The algorithm flow of the active dynamic motion decoupling and disturbance suppression algorithm is as follows:

[0037] a. Establish the dynamic model of the entire detection system: First, establish dynamic equations including mass, damping, and stiffness for the entire circular track patrol platform, adaptive image-stabilized robotic arm, and secondary active stabilization gimbal:

[0038]

[0039] Where q is the generalized coordinate representing the position of each joint; M(q) is the system inertia matrix; G(q) represents the Coriolis force and centrifugal force matrix; G(q) is the gravity term, τ is the control torque, and τd is the external disturbance.

[0040] b. Dynamic motion decoupling: The control law is designed as follows:

[0041]

[0042] Where u is the new control input;

[0043] Substituting the above control law into the original dynamic equation, we obtain the decoupled linear system:

[0044]

[0045] Under the control of the edge computing module, the drive system actively outputs a torque τ;

[0046] Further design of u to track the desired trajectory and suppress disturbances;

[0047] c. Disturbance Observer Design: Define the estimated value of the disturbance observer as... Its renewal law is:

[0048]

[0049] It compares the actual position q of the adaptively stabilized robotic arm with the position it should reach in an ideal, undisturbed environment. The differences between To estimate external vibration disturbances in real time. Size and orientation;

[0050] Where L is the gain matrix of the perturbation observer, The velocity estimate is calculated using a dynamic model; the following perturbation observer structure is adopted:

[0051]

[0052] Among them, K o It is the correction gain of the disturbance observer;

[0053] d. Feedforward-feedback composite control, employing a feedforward-feedback control structure; the feedback control uses PD control, and the feedforward control utilizes the output of the disturbance observer; the control input u is designed as: .

[0054] Among them, K p and K d These are control parameters;

[0055] An adaptive notch filter algorithm is introduced. The transfer function of the notch filter algorithm is:

[0056]

[0057] Where, ω n ζwidth and ζdepth are the notch frequency, which is the vibration frequency to be suppressed; ζwidth and ζdepth are the damping coefficients.

[0058] When the segment assembly machine moves: the active dynamic motion decoupling and disturbance suppression algorithm estimates the torque disturbance caused by the movement of the segment assembly machine through the disturbance observer and cancels it out through feedforward compensation. At the same time, the notch filter suppresses the main vibration frequency to ensure the stability of the end of the robotic arm. When the tunnel boring machine advances: the low-frequency vibration generated by the advance is suppressed by feedback control and feedforward compensation, while the high-frequency vibration is handled by the secondary active stabilization gimbal.

[0059] The active dynamic motion decoupling and disturbance suppression algorithm continuously receives acceleration and angular velocity data from the IMU sensor network deployed in each arm joint of the robot body base, main body and end effector; it performs optimal fusion through Kalman filter to accurately estimate the amplitude, frequency and phase of vibration; based on this, it generates feedforward compensation signal, which is directly superimposed on the current loop of the servo motor of each joint;

[0060] When the segment assembly machine moves the segment to be installed: the active dynamic motion decoupling and disturbance suppression algorithm controls the patrol platform base of the ring track to perform smooth tracking motion. At the same time, the adaptive image stabilizing robotic arm and the secondary active stabilizing gimbal work together to counteract the low-frequency shaking and high-frequency jitter caused by the start and stop of the patrol platform base and the movement of the segment assembly machine, so that the global array camera can always obtain clear motion images of the segment to be installed.

[0061] The algorithm flow for the segment misalignment detection algorithm is as follows:

[0062] a. Rapid macroscopic identification and positioning:

[0063] Deep learning-based semantic segmentation of seams identifies seams using a convolutional network with an encoder-decoder structure.

[0064]

[0065] The loss function uses a combination of cross-entropy and Dice loss:

[0066]

[0067] Where y is the actual seam location. It is the joint location predicted by the segment misalignment detection algorithm;

[0068] The transformation matrix is ​​obtained through hand-eye calibration. Convert the image coordinates to the robot arm's base coordinate system:

[0069]

[0070] Using a pre-calibrated transformation matrix This involves accurately converting a pixel with coordinates (u, v) and depth d on an image to its three-dimensional coordinates (X, v) in the robot arm's base coordinate system. b ,Y b Z b This allows the system to obtain the specific location of the seam seen in the global array camera in real space, and then control the adaptive image stabilization robotic arm to move to that specific location in real space.

[0071] b. Visual servoing guidance:

[0072] Design image-based visual servo control:

[0073] Where s represents the image feature. For desired features;

[0074] The control law is:

[0075] Where e is the segment feature in the image, i.e., the deviation between the current position and the desired position. It is the pseudo-inverse of a matrix that reflects the relationship between the motion of a global array camera and image changes;

[0076] c. Precise extraction of microscopic geometric features:

[0077] A robust ICP variant algorithm is used for point cloud registration and model matching:

[0078]

[0079] Wherein, weight w i Based on point-to-point distance and normal vector similarity calculation; through an iterative nearest-point algorithm, a set of rotations R and translations t is found such that the scanned point cloud p... i With design model q c(i) Best match;

[0080] Robust least squares was used for plane fitting on both sides of the joint.

[0081]

[0082] Where ρ is the Huber loss function;

[0083] Calculation of misalignment:

[0084]

[0085] On the registered point cloud, two planes are fitted using point cloud on both sides of the seam. and The misalignment δ is calculated by taking the average difference between the distances from all points to the opposite plane.

[0086] d. Analysis of joint geometric characteristics:

[0087] Use standard deviation analysis to analyze width consistency:

[0088]

[0089] Among them, w i It is a single width value measured at different locations along the seam; It is the arithmetic mean of the widths of all measurement points, i.e., the "average width"; σ w The smaller the value of σ, the closer the width values ​​of all measurement points are to the average value, and the more uniform the joint width; conversely, the larger the value of σ, the more uniform the joint width. w The larger the value, the more uneven the width of the joint, indicating poor uniformity; if σ w If the threshold is exceeded, the segment misalignment detection algorithm determines that the width of the joint is inconsistent.

[0090] e. Straightness error analysis:

[0091]

[0092] The segment misalignment detection algorithm uses the least squares method to analyze all 3D point cloud data points p on the joint centerline. i Fit an optimal three-dimensional straight line L fit Then, calculate the distance from each point on the center line to the fitted straight line L. fit vertical distance dist(p) i ,L fit Finally, the maximum value among all these distances is taken as the straightness error ε. straight , ε straight The smaller the value, the closer the center line of the joint is to a perfect straight line; the larger the value, the more obvious the bend or serpentine twist of the joint.

[0093] f. Coplanarity assessment:

[0094]

[0095] First, the least squares method is used to fit an optimal plane to the point cloud data of the segment surface on one side of the joint; the mathematical expression of this plane is n·X + d = (0), where n is the unit normal vector of the plane and d is a constant term; for a point p belonging to this side surface i Substituting the absolute value obtained into the plane equation |n·p i + d|, geometrically, represents the perpendicular distance from the point to the fitted plane; ε flat That is, the maximum distance from all points to the plane, used to assess whether the tunnel inner wall formed after the segments to be installed are smooth and flat; ε flat The smaller the value, the smoother the surface of the segment; the larger the value, the more likely there is a local bulge or depression.

[0096] Compared with the prior art, the present invention has the following advantages:

[0097] 1. This invention has strong adaptability and is easy to install and remove: The detection system is installed on the rear side of the H-beam at the tail of the tunnel boring machine through a detachable connection structure, which does not take up space, does not require a separate support frame, is compatible with different models of tunnel boring machines, and facilitates the installation, disassembly and maintenance of the detection system, thereby reducing construction preparation costs.

[0098] 2. This invention has a wide detection range and high flexibility: the robotic arm body can move 360° in a circular motion along the circular guide rail. Combined with the multi-degree-of-freedom telescopic design, it can cover the detection needs of segment misalignment joints in different circumferences and distances in the tunnel, leaving no blind spots and solving the problem of limited detection range of fixed detection equipment.

[0099] 3. This invention is resistant to vibration interference and has high detection accuracy: The two-stage active stabilizing gimbal adopts a dual-axis stabilization structure, combined with real-time vibration monitoring and compensation technology. Through the dual anti-vibration design of "physical isolation + active algorithm decoupling", it can effectively offset the vibration generated by the operation of surrounding equipment during tunnel construction, avoid detection errors caused by vibration, and control the detection error within 0.1mm, far exceeding that of fixed equipment. This ensures the accuracy of segment misalignment detection data, provides a reliable basis for tunnel construction quality control, and can meet the most stringent construction specifications.

[0100] 4. This invention has a high degree of automation and good safety: The detection system can work in conjunction with the shield machine control system and the segment assembly machine to realize the automated control of the detection process. No staff need to enter the tunnel to work, which reduces the labor intensity and safety risks of manual detection. Single-point detection only takes 30 seconds, which improves detection efficiency and truly realizes "assembly and detection at the same time", providing instant feedback for construction decisions. Attached Figure Description

[0101] Figure 1This is a schematic diagram of the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm, according to the present invention.

[0102] Figure 2 This is an installation diagram of the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm, according to the present invention.

[0103] Figure 3 This is the installation front view of the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm according to the present invention;

[0104] Figure 4 This is a schematic diagram of the composition structure of the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm according to the present invention;

[0105] Figure 5 This is a partial enlarged view of the ring track patrol platform in the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm, as described in this invention.

[0106] Figure 6 This is a schematic diagram of the adaptive image-stabilized robotic arm in the high-precision inspection system for tunnel segment joint quality based on the image-stabilized robotic arm of the present invention.

[0107] Figure 7 This is a schematic diagram of the macro-micro fusion detection module in the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm, as described in this invention.

[0108] Figure 8 This is a schematic diagram of the tunnel segment annular joint inspection system based on a stabilized robotic arm for high-precision inspection of tunnel segment joint quality according to the present invention.

[0109] Figure 9 This is a schematic diagram of the adjacent segment joint detection of the high-precision tunnel segment joint quality detection system based on a stabilized robotic arm according to the present invention.

[0110] Figure 10 This is a schematic diagram of the high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm, according to the present invention.

[0111] Figure 11 This is a flowchart of the high-precision inspection method for tunnel segment joint quality based on a stabilized robotic arm, according to the present invention.

[0112] In the diagram, the components are: 1. Ring-rail patrol platform; 1-1. Patrol platform base; 1-2. Rail guide roller; 1-3. Ring-rail base; 1-4. Guide slide rail; 1-5. Patrol drive system; 1-6. Precision geared rail; 2. Adaptive image stabilization robotic arm; 2-1. End effector interface gripper; 2-2. Robotic arm body; 2-3. Adaptive edge computing module; 3. Macro-micro fusion detection module; 3. Secondary active stabilization gimbal; 3-1. Global array camera; 3-2. 3D infrared scanner; 3-3. H-beam; 4. Vibration damping bolt; 5. High-strength connector; 7. Vibration damping connector; 8. Segment to be installed; 9-1. Installed segment; 9-2. Segment assembly machine; 9-3. Detailed Implementation

[0113] The present invention will be further described below with reference to the accompanying drawings and specific embodiments.

[0114] Please see the appendix Figure 1 To be continued Figure 4 A high-precision inspection system for tunnel segment joint quality based on a stabilized robotic arm includes a ring-rail patrol platform 1, an adaptive stabilized robotic arm 2, and a macro-micro fusion inspection module 3. The ring-rail patrol platform 1 includes a patrol platform base 1-1 and a guide rail. The guide rail is installed on the H-beam 4 at the tail of the tunnel boring machine via a shock-absorbing connector 8. The guide rail has a circular structure and is located inside the tunnel boring machine. The patrol platform base 1-1 is set on the guide rail and can move circumferentially along the guide rail. One end of the adaptive stabilized robotic arm 2 is installed on the patrol platform base 1-1, and the macro-micro fusion inspection module 3 is installed on the other end of the adaptive stabilized robotic arm 2, so that the macro-micro fusion inspection module 3 can move synchronously with the patrol platform base 1-1 and face the joint between the segment to be installed 9-1 and the installed segment 9-2.

[0115] The circular track patrol platform 1 serves as the core of the entire detection system for both motion and load-bearing. The adaptive image-stabilized robotic arm 2 is an intelligent electromechanical system integrating sensing, control, and execution. Its function is to achieve high-precision motion while actively suppressing external vibrations, providing a stable detection environment for the macro-micro fusion detection module 3. The macro-micro fusion detection module 3 is an intelligent sensing terminal integrating active image stabilization, macroscopic detection, microscopic measurement, and communication feedback, serving as the terminal directly executing the detection tasks.

[0116] The entire detection system is mounted on the H-beam 4 at the tail of the tunnel boring machine via a ring-rail patrol platform 1, which makes the entire detection system "suspended" outside the tunnel construction environment, thereby minimizing the impact of construction vibration on the accuracy of the measurement.

[0117] Please see the appendix Figure 5The guide track includes a rail-clamping guide roller 1-2, a ring rail base 1-3, a guide slide rail 1-4, a cruise drive system 1-5, and a precision geared rail 1-6. The guide slide rail 1-4 and the precision geared rail 1-6 are both circular ring structures and are concentrically stacked on the ring rail base 1-3 and fixed by a high-strength connector 7. The ring rail base 1-3 is installed on the H-beam 4 at the tail of the tunnel boring machine through a shock-absorbing connector 8. A pair of rail-clamping guide rollers 1-2 are spaced below the cruise platform base 1-1, and the pair of rail-clamping guide rollers 1-2 are respectively matched and rolled in engagement with the inner and outer rings of the guide slide rail 1-4. The cruise drive system 1-5 is installed on the cruise platform base 1-1 and is meshed and driven by the precision geared rail 1-6. The adaptive image-stabilized robotic arm 2 is installed on the cruise platform base 1-1.

[0118] The working principle of the circum-rail cruise platform 1 is as follows:

[0119] When the servo motor of the cruise drive system 1-5 receives a command from the adaptive edge computing module 2-3, it begins to rotate. The torque is amplified by the reducer and transmitted to the drive gear. Since the precision gear rail 1-6 is stationary, according to the principle of action and reaction, the rotating drive gear applies a tangential force to the precision gear rail 1-6. This force, in turn, pushes the drive gear itself and the entire cruise platform base 1-1, which is rigidly connected to it, to move along the circumference of the precision gear rail 1-6. Simultaneously with the output power of the cruise drive system 1-5, the guide roller 1-2 begins to operate. The guide roller 1-2 rolls purely on the fixed guide rail 1-4. Through its precise groove and engagement with the guide rail 1-4, the radial and axial degrees of freedom of the cruise platform base 1-1 are strictly restricted, retaining only one degree of freedom: rotation around the center of the guide rail 1-4. The opposing guide rollers 1-2 eliminate the gap between themselves and the guide rails 1-4 through preloading, ensuring that the cruise platform base 1-1 is firmly constrained at any position. This achieves a sway-free, high-rigidity "cruising" motion and effectively resists the overturning moment caused by load eccentricity. Thanks to the extremely low frictional resistance of the guide rollers 1-2 rolling on the guide rails 1-4 and the smooth speed regulation of the servo motor, the cruise platform base 1-1 can move smoothly and silkily.

[0120] Ultimately, the ring-track patrol platform 1 enables the onboard adaptive image-stabilized robotic arm 2 and macro-micro fusion detection module 3 to cover the entire annular area of ​​the tunnel boring machine tail, achieving precise detection operations without blind spots.

[0121] The cruise platform base 1-1 serves as the mounting base for the adaptive image-stabilized robotic arm 2. It also acts as a common carrier for all moving parts, bearing the entire weight of the adaptive image-stabilized robotic arm 2 and the macro-micro fusion detection module 3, and connecting the circular cruise platform 1 and the adaptive image-stabilized robotic arm 2 into a single unit. Preferably, the cruise platform base 1-1 is precision-machined from lightweight, high-strength aluminum alloy, exhibiting high rigidity and lightweight characteristics. Its upper surface is fitted with loads such as the adaptive image-stabilized robotic arm 2 via vibration-damping bolts, and also integrates the mounting position for the cruise drive system 1-5.

[0122] Preferably, the guide roller 1-2 can be composed of four high-lubrication groove bearings arranged in opposite directions. The grooves are precision ground to form a high-precision conformal match with the arc-shaped contour of the guide rail 1-4. The guide roller assembly 1-2 engages with the guide rail 1-4, realizing smooth annular guidance and stable support for the cruise platform pedestal 1-1. It bears the entire vertical load of the cruise platform pedestal 1-1, effectively resists overturning moment, and ensures that the cruise platform pedestal 1-1 will not shake, jump, or derail during movement.

[0123] Preferably, the cruise drive system 1-5 consists of a servo motor, a reducer, and a drive gear. The drive gear is directly mounted on the output shaft of the reducer. The cruise drive system 1-5 is rigidly mounted on the mounting position of the cruise platform base 1-1, ensuring a stable meshing center distance between the drive gear and the precision geared rail 1-6. The drive gear meshes with the fixed precision geared rail 1-6, and under the power output of the servo motor and the precise control of the reducer, the rotation of the drive gear is converted into a precise and smooth "cruising" motion along the circular precision geared rail 1-6.

[0124] Preferably, the ring rail base 1-3 can be made of high-strength lightweight alloy material and is circular in shape. Its inner diameter is adapted to the outer diameter of the H-beam 4 at the tail of the tunnel boring machine. The ring rail base 1-3 is installed on the rear side of the H-beam 4 at the tail of the tunnel boring machine through the shock-absorbing connector 8. Elastic buffer pads are provided at the connection structure to reduce the vibration transmitted to the ring rail base 1-3 during the operation of the tunnel boring machine. At the same time, it facilitates the installation and disassembly of the entire detection system and adapts to the needs of different tunnel boring machines.

[0125] Preferably, the guide rail 1-4 is a high-precision ring rail made of high-hardness bearing steel, with a surface that has undergone heat treatment and grinding, resulting in extremely low roughness and high wear resistance. The guide rail 1-4 is stacked and fixed on the ring rail base 1-3 with the precision gear rail 1-6 in a coaxial and coplanar manner via a high-strength connector 7, providing a smooth and rigid running track for the guide roller 1-2.

[0126] Preferably, the precision meshing gear 1-6 can be a high-precision ring rack with precision-machined teeth to ensure smooth, low-noise, and backlash-free engagement with the drive gear of the cruise drive system 1-5.

[0127] Preferably, the shock-absorbing connector 8 can be made of high-strength shock-absorbing bolts, and the high-strength connector 7 can be made of high-strength bolts.

[0128] Please see the appendix Figure 6 The adaptive image-stabilized robotic arm 2 includes an end effector interface gripper 2-1, a robotic arm body 2-2, and an adaptive edge computing module 2-3. The base of the robotic arm body 2-2 is mounted on the cruise platform 1-1 of the circular track cruise platform 1. The robotic arm body 2-2 is composed of several arm joints connected in series by servo motors, and each joint embeds an inertial measurement unit (IMU). The adaptive edge computing module 2-3 is mounted on the robotic arm body 2-2 and connected to the robotic arm body 2-2 through a data transmission cable 6. The adaptive edge computing module 2-3 integrates a segment misalignment detection algorithm and an active dynamic motion decoupling and disturbance suppression algorithm. The end effector interface gripper 2-1 is mounted at the end of the robotic arm body 2-2 away from the cruise platform 1-1. The end effector interface gripper 2-1 integrates a pitch joint and a rotation joint. The macro-micro fusion detection module 3 is mounted at the end of the robotic arm body 2-2 through the pitch joint and the rotation joint.

[0129] On the large-scale, high-precision circular motion bearing benchmark achieved by the circular track patrol platform 1, the adaptive image stabilization robotic arm 2 further adjusts the pose of the macro-micro fusion detection module 3 installed at the end through the movement of multiple arm joints of its robotic arm body 2-2, realizing multi-angle and complex position detection, which greatly expands the flexibility and detection range of the entire detection system.

[0130] Preferably, the robotic arm body 2-2 is composed of six arm joints connected in series, but the number of arm joints connected in series can be adaptively adjusted according to actual usage requirements. The robotic arm body 2-2 can be made of carbon fiber or aerospace aluminum alloy. Through topology optimization design, it achieves lightweight while ensuring extremely high rigidity, thereby reducing motion inertia and improving response speed. The base of the robotic arm body 2-2 can be mounted on the cruise platform base 1-1 via shock-absorbing bolts 5.

[0131] Preferably, each arm joint consists of a high-precision servo motor, a harmonic reducer, and a high-resolution encoder, thereby enabling the end effector of the robotic arm body 2-2 to achieve high degree of freedom and high-precision pose adjustment in space. At the same time, each joint is embedded with an IMU sensor to collect vibration data and feed it back to the adaptive edge computing module 2-3.

[0132] The adaptive edge computing module 2-3 serves as the control center of the entire detection system. Preferably, the adaptive edge computing module 2-3 can adopt a high-performance embedded computer with existing technology and integrate it on the robotic arm body 2-2, including a multi-core CPU and a high-performance AI graphics card. Internally, it runs the segment misalignment detection algorithm and the active dynamic motion decoupling and disturbance suppression algorithm.

[0133] The pitch and rotation joints integrated within the end effector interface gripper 2-1 serve as precision dual-axis image stabilizing joints providing two active degrees of freedom of motion. They can precisely control the macro-micro fusion detection module 3 to achieve angle rotation and pitch adjustment.

[0134] Through the active dynamic motion decoupling and disturbance suppression algorithms of the shock-absorbing bolts 5 and the adaptive image-stabilized robotic arm 2, complex vibrations from the tunnel boring machine and the ring track patrol platform 1 can be effectively isolated and compensated, providing an extremely stable working environment for the macro-micro fusion detection module 3 at the end, ensuring clear and accurate data acquisition.

[0135] Please see the appendix Figure 7 The macro-micro fusion detection module 3 includes a secondary active stabilizing gimbal 3-1, a global array camera 3-2, and a three-dimensional infrared scanner 3-3. The secondary active stabilizing gimbal 3-1 is detachably connected to the end effector interface gripper 2-1 through a self-locking mechanism. The global array camera 3-2 and the three-dimensional infrared scanner 3-3 are mounted on the secondary active stabilizing gimbal 3-1 through a quick-release structure. The global array camera 3-2 and the three-dimensional infrared scanner 3-3 are positioned facing the joint between the segment to be installed 9-1 and the installed segment 9-2.

[0136] Preferably, the self-locking mechanism is built into the end effector interface gripper 2-1, used to connect the secondary active stabilizing gimbal 3-1 and the end effector interface gripper 2-1. The self-locking mechanism allows for one-click quick assembly and disassembly of the macro-micro fusion detection module 3, while ensuring connection rigidity and accuracy repeatability. The self-locking mechanism integrates a power supply and a high-speed data communication interface, providing power to the macro-micro fusion detection module 3 and enabling lossless, low-latency data transmission. The self-locking mechanism is a commonly used connection mechanism in this field, and its specifications and models can be adaptively selected according to the actual connection requirements of the macro-micro fusion detection module 3.

[0137] Preferably, the secondary active stabilizing gimbal 3-1 can be a dual-axis precision servo platform, which integrates a high-performance IMU for directly and quickly sensing residual high-frequency vibrations of the macro-micro fusion detection module 3. It also embeds a collaborative control and communication unit responsible for communicating with the edge computing module. The secondary active stabilizing gimbal 3-1 and the adaptive image-stabilizing robotic arm 2 constitute a "master-slave" two-stage image stabilization system. The robotic arm body 2-2 compensates for large-amplitude, low-frequency vibrations, while the secondary active stabilizing gimbal 3-1 is responsible for counteracting residual, higher-frequency micro-shakes, providing pixel-level stabilization for the global array camera 3-2 and the 3D infrared scanner 3-3.

[0138] Preferably, the global array camera 3-2 can be composed of multiple high-resolution, high-frame-rate global shutter CMOS sensors to form an ultra-wide field of view for rapid coverage of large areas. The macro-micro fusion detection module 3 is surrounded by multi-angle LED supplementary lights to ensure uniform and clear macroscopic images even in low-light environments. The 3D infrared scanner 3-3 employs structured light technology and includes precise emitting and receiving lens groups to acquire high-precision 3D point cloud data of the object's surface, ensuring the accuracy of the measurement optical path.

[0139] Please see the appendix Figure 8 To be continued Figure 10 The working principle of the entire detection system is as follows: The circular track patrol platform 1 transports the adaptive image-stabilized robotic arm 2 to the approximate installation area of ​​the segment 9-1 to be installed. Upon receiving the detection command from the adaptive edge computing module 2-3, the macro-micro fusion detection module 3 is activated. The collaborative control and communication unit built into the macro-micro fusion detection module 3 first performs a self-test to confirm that the secondary active stabilization gimbal 3-1, the global array camera 3-2, and the 3D infrared scanner 3-3 are in normal working order. The secondary active stabilization gimbal 3-1 activates its built-in high-performance IMU and enters standby mode, ready to perform the image stabilization task.

[0140] According to the system plan, the adaptive edge computing module 2-3 issues a "execute macro scan" command to the macro-micro fusion detection module 3. During the movement or stillness of the robotic arm body 2-2, the secondary active stabilizing gimbal 3-1 continuously operates. Based on high-frequency vibration data collected by its own IMU, it drives its internal motor for reverse compensation, ensuring a stable reference for the global array camera 3-2 and the 3D infrared scanner 3-3. Subsequently, the global array camera 3-2 starts up, performing high-speed, multi-angle imaging of the designated tunnel segment area to acquire high-resolution panoramic image data. The collaborative control and communication unit inside the macro-micro fusion detection module 3 standardizes and packets the captured panoramic image data, attaching timestamps, module IDs, and other information. Then, through a high-speed data interface, the raw data is transmitted losslessly to the adaptive edge computing module 2-3.

[0141] The adaptive edge computing modules 2-3 receive various types of data collected by the macro-micro fusion detection module 3. At this point, the segment misalignment detection algorithm takes over: it first processes the macroscopic image from the macro-micro fusion detection module 3, identifies the joint to be detected through a deep learning model, and sends a "perform microscopic measurement" command to the macro-micro fusion detection module 3, along with the precise coordinates of the target joint area. Then, through the hand-eye calibration matrix, it converts the image coordinates into a three-dimensional target pose in the robotic arm's base coordinate system.

[0142] Based on the target pose in three-dimensional space, the active dynamic motion decoupling and disturbance suppression algorithm is activated to perform motion planning. According to the dynamic model of the detection system, it calculates the optimal trajectory of each arm joint and the pitch and rotation joints of the end effector of the circular track patrol platform 1 and the robotic arm body 2-2, ensuring that the motion of each component is smooth and free from coupling interference.

[0143] The adaptive edge computing module 2-3 sends the planned trajectory commands to the high-precision servo motors of each arm joint. Under the feedback of their high-resolution encoders, each arm joint of the robotic arm body 2-2 begins to move collaboratively, driving the macro-micro fusion detection module 3 at its end effector to move towards the target pose. Throughout the movement, the IMU sensor network distributed across the base, main body, and end effector of the robotic arm body 2-2 continuously collects six-degree-of-freedom vibration data. The disturbance observer in the active dynamic motion decoupling and disturbance suppression algorithm runs in real time, estimating external disturbances from the tunnel boring machine and the circular track patrol platform 1 based on the adaptive edge computing module 2-3. The active dynamic motion decoupling and disturbance suppression algorithm uses this estimated external disturbance as a feedforward quantity, synthesizing it with the feedback control quantity based on position error to generate the final arm joint control torque. This arm joint control torque directly cancels out external vibrations, ensuring that the end effector of the robotic arm body 2-2 remains stable during movement. During this process, the segment misalignment detection algorithm continuously runs, processing the real-time images acquired by the global array camera 3-2 and calculating the image deviation between the current joint features and the desired position. Through visual servo control laws, fine-tuning instructions are generated to dynamically correct the motion trajectories of each joint of the robotic arm 2-2, enabling the robotic arm 2-2 to move flexibly within its degrees of freedom, ultimately achieving precise "eye-to-hand" locking.

[0144] When the end effector of the robotic arm 2-2 reaches near the target pose, the system enters a "high-precision locking mode." At this time, the main movements of the upper arm joints of the robotic arm 2-2 tend to stop, while the pitch and rotation joints of the end effector interface gripper 2-1 drive the macro-micro fusion detection module 3 to perform fine-tuning of the viewing angle. Simultaneously, based on the high-performance IMU data integrated into the connected secondary active stabilizing gimbal 3-1, millisecond-level micro-motion compensation is performed for residual, higher-frequency micro-vibrations. This constitutes a master-slave collaborative image stabilization system in which "the robotic arm 2-2 resists large-amplitude low-frequency vibrations, and the connected secondary active stabilizing gimbal 3-1 cancels small-amplitude high-frequency vibrations," ultimately providing a pixel-level stable observation platform for the macro-micro fusion detection module 3.

[0145] The 3D infrared scanner 3-3 projects coded structured light onto the target seam area according to instructions, and simultaneously receives the deformed light spots modulated by the surfaces of the segment to be installed 9-1 and the installed segment 9-2, thereby directly generating high-density, high-precision 3D point cloud data of the target seam area. Similarly, the collaborative control and communication unit packages the original 3D point cloud data, attaches the necessary metadata, and uploads it completely to the adaptive edge computing module 2-3 via a high-speed link.

[0146] Please see the appendix Figure 1 To be continued Figure 11 A macro-micro fusion detection method for tunnel segment joint quality based on a stabilized robotic arm includes the following steps:

[0147] Step 1: Position the segment 9-1 to be installed, that is, transport the segment 9-1 to the assembly position.

[0148] Step 2: The detection system receives the detection command, and the circular track patrol platform 1 drives the adaptive image stabilizing robotic arm 2 to move, monitoring the placement process of the segment 9-1 to be installed.

[0149] Step 2 includes the following sub-steps:

[0150] Step 21: When the segment 9-1 to be installed is transported to the assembly position, the circular track patrol platform 1, under the control of the adaptive edge computing module 2-3, drives the adaptive image-stabilized robotic arm 2 and the macro-micro fusion detection module 3 to move to the predetermined observation position. The predetermined observation position can be determined according to the assembly position of the segment 9-1 to be installed, ensuring that the macro-micro fusion detection module 3 at the predetermined observation position can acquire the required images to monitor the placement process of the segment 9-1 to be installed.

[0151] Step 22: The adaptive image stabilization robotic arm 2 maintains stable tracking during the movement of the tube segment 9-1 to be installed through an active dynamic motion decoupling and disturbance suppression algorithm.

[0152] Step 23: The global array camera 3-2 of the macro-micro fusion detection module 3, under the image stabilization guarantee of the secondary active stabilizing gimbal 3-1, continuously captures images of the moving tube segment 9-1 to be installed, and transmits the real-time image data stream to the adaptive edge computing module 2-3 to realize the monitoring of the placement process of the tube segment 9-1 to be installed.

[0153] Step 3: Assemble the segment 9-1 to be installed using the segment assembly machine 9-3.

[0154] Step 4: The circular track patrol platform 1 drives the adaptive image stabilization robotic arm 2 to move along the guide rails 1-4. The adaptive image stabilization robotic arm 2 controls the macro-micro fusion detection module 3 at its end to move to the target detection area.

[0155] Step 5: The macro-micro fusion detection module 3 is assembled and positioned to conduct preliminary detection of the target detection area at the joint of the segment to be installed 9-1.

[0156] Specifically, the macro-micro fusion detection module 3 performs a rapid macro-scan of the circumferential and tangential seams of the segment 9-1 to be installed: the global array camera 3-2 acquires panoramic images and transmits the panoramic image data in real time to the adaptive edge computing module 2-3 for preliminary seam identification, and records the macro position and orientation data of the segment 9-1 to be installed.

[0157] Step 6: Based on the preliminary detection results, precise positioning and image stabilization are performed using the adaptive image stabilization robotic arm 2 and the macro-micro fusion detection module 3.

[0158] Specifically, based on the preliminary detection results in step 5, namely the macroscopic position and attitude data of the segment 9-1 to be installed, the fine detection stage begins: the adaptive edge computing module 2-3 calculates the precise three-dimensional coordinates of the joint through the hand-eye calibration matrix, the ring-track patrol platform 1 performs coarse positioning, the adaptive image-stabilized robotic arm 2 is moved to the target area, the robotic arm body 2-2 of the adaptive image-stabilized robotic arm 2 performs fine pose adjustment through the movement of each arm joint, and performs micro-angle compensation through the pitch and rotation joints built into the end effector interface gripper 2-1, and the secondary active stabilization gimbal 3-1 activates high-frequency vibration suppression to achieve pixel-level stabilization.

[0159] Step 7: The macro-micro fusion detection module 3 performs microscopic measurements and data acquisition on the target detection area at the joint.

[0160] Specifically, the adaptive edge computing module 2-3 sends measurement commands to the macro-micro fusion detection module 3, the three-dimensional infrared scanner 3-3 performs high-density point cloud acquisition on the target joint, and transmits all the original three-dimensional point cloud data to the adaptive edge computing module 2-3 without loss. The segment misalignment detection algorithm processes the data in real time and calculates parameters such as misalignment amount and joint width.

[0161] Step 8: Inspect the entire ring of segments 9-1 to be installed in sequence according to the segment assembly order. After completing the joint inspection of one ring of segments, repeat steps 1-7 and generate a quality inspection report for each joint in real time.

[0162] Specifically, after completing the joint inspection of one ring, the entire inspection system moves with the tunnel boring machine to the assembly position of the next ring segment and automatically plans the joint inspection path of the next ring segment. The patrol platform base 1-1 moves precisely along the guide rail 1-4 to the next inspection position.

[0163] Step 9: After completing the inspection of all the segments 9-1 to be installed, the adaptive stabilizing robotic arm 2 retracts each joint to the transport posture, and the patrol platform base 1-1 returns to the initial position along the guide rail 1-4.

[0164] At this point, the entire detection system enters a low-power standby state, maintaining basic image stabilization, and all detection data is archived and a comprehensive quality assessment report is generated.

[0165] When a new segment is transported to the assembly position, the detection system automatically wakes up and initializes each sensor, updates the detection path according to the new segment assembly plan, and repeats the above complete detection process, i.e., steps 1 to 8.

[0166] The robotic arm body 2-2 and the secondary active stabilizing gimbal 3-1 mounted on it can reduce mutual interference through mechanical decoupling using an active dynamic motion decoupling and disturbance suppression algorithm. The algorithm flow of the active dynamic motion decoupling and disturbance suppression algorithm is as follows:

[0167] a. Establish the dynamic model of the entire detection system: First, establish dynamic equations including mass, damping, and stiffness for the entire circular track patrol platform 1, adaptive image-stabilized robotic arm 2, and secondary active stabilized gimbal 3-1:

[0168]

[0169] Where q is a generalized coordinate representing the position of each joint (including each arm joint, pitch joint, and rotation joint); M(q) is the system inertia matrix. G(q) represents the Coriolis force and centrifugal force matrix; G(q) is the gravity term, τ is the control torque, and τd is the external disturbance.

[0170] Based on the above dynamic equations, the adaptive edge computing module 2-3 predicts and simulates how the system will move after being subjected to force, which is the cornerstone of all advanced control strategies.

[0171] b. Dynamic Motion Decoupling: To eliminate the coupling effects between different motion axes, feedback linearization is performed using the torque calculation method. The control law is designed as follows:

[0172]

[0173] Where u is the new control input. Substituting the above control law into the original dynamic equations, we obtain the decoupled linear system:

[0174]

[0175] Under the control of the edge computing module, the drive system actively outputs a torque τ to precisely counteract the internal coupling forces such as gravity and inertia within the system itself. This transforms the three originally tightly coupled motion units—the cruise platform base 1-1, the adaptive image-stabilized robotic arm 2, and the secondary active stabilization gimbal 3-1—into three relatively independent linear systems. The effect is that when the cruise platform base 1-1 is commanded to rotate over a large range, the adaptive image-stabilized robotic arm 2 and the secondary active stabilization gimbal 3-1 will not be "led astray" by the movement of the cruise platform base 1-1, thus achieving coordinated movement where each performs its function without interference.

[0176] At this point, the detection system is transformed into a linear second-order system, but the perturbation term still exists. Further design is needed to u to track the desired trajectory and suppress the perturbation.

[0177] c. Disturbance observer design: To estimate and compensate for external vibration disturbances. Design a disturbance observer. Define the estimated value of the disturbance observer as follows: Its renewal law is:

[0178]

[0179] It compares the "actual position" q of the adaptive image-stabilized robotic arm 2 with the "position it should be in an ideal, undisturbed environment". The differences between To estimate external vibration disturbances in real time. Size and orientation.

[0180] Where L is the gain matrix of the perturbation observer, The velocity estimate is calculated using a dynamic model. The following perturbation observer structure is employed:

[0181]

[0182] Among them, K o This is the correction gain of the disturbance observer. By selecting appropriate L and K... o This ensures the convergence of the perturbation estimation.

[0183] d. Feedforward-feedback composite control, employing a feedforward-feedback control structure. Feedback control uses PD control, while feedforward control utilizes the output of a disturbance observer. The control input u is designed as follows: .

[0184] This includes the concept of "target trajectory". "Current error" (q) d -q) and the "early warning information" from the disturbance observer, i.e., the estimated value. K p and K d These are control parameters, and their design ensures that the energy function (Lyapunov function V) of the entire detection system always decreases, i.e. This mathematically guarantees that, regardless of the magnitude of the disturbance, the dynamic model can automatically and stably converge to the target position without going out of control or oscillating.

[0185] To address the main frequency components of tunnel boring machine (TBM) vibration, we introduce an adaptive notch filter algorithm into the control loop. The transfer function of the notch filter algorithm is:

[0186]

[0187] Where, ω n ζwidth and ζdepth are the notch frequency, representing the vibration frequency to be suppressed; ζwidth and ζdepth are the damping coefficients. ωwidth is adaptively adjusted by analyzing the vibration spectrum in real time. n This system tracks changes in vibration frequency and selectively suppresses the main frequency components of the tunnel boring machine's vibration. Simultaneously, by using an IMU sensor network deployed across various parts of the robotic arm, combined with Kalman filters for data fusion, the vibration spectrum is accurately identified, and control parameters are dynamically adjusted to achieve precise frequency tracking and suppression.

[0188] When the segment assembler 9-3 moves: the active dynamic motion decoupling and disturbance suppression algorithm estimates the torque disturbance caused by the movement of the segment assembler 9-3 through a disturbance observer, and cancels it out through feedforward compensation. At the same time, a notch filter suppresses the main vibration frequency, ensuring the stability of the end of the robotic arm 2-2. When the tunnel boring machine advances: the low-frequency vibration generated by the advance is suppressed by feedback control and feedforward compensation, while the high-frequency vibration is handled by the secondary active stabilizing gimbal 3-1 (using a similar algorithm). The secondary active stabilizing gimbal 3-1 has its own stabilization and vibration resistance algorithm, and its process for handling high-frequency vibration will not be described in detail here.

[0189] The adaptive edge computing module 2-3 incorporates an active dynamic motion decoupling and disturbance suppression algorithm to continuously receive acceleration and angular velocity data from the IMU sensor network deployed in the base, main body, and end effector of the robotic arm 2-2. Optimal fusion is performed using a Kalman filter to accurately estimate the amplitude, frequency, and phase of the vibration. Based on this, a feedforward compensation signal is generated and directly superimposed on the current loop of the servo motors in each joint, achieving a pre-emptive compensation effect for vibration, significantly improving the system's response speed and steady-state accuracy.

[0190] When the segment assembly machine 9-3 moves the segment 9-1 to be installed: the active dynamic motion decoupling and disturbance suppression algorithm controls the patrol platform base 1-1 of the circular track patrol platform 1 to perform smooth tracking motion. At the same time, the adaptive image stabilizing robotic arm 2 and the secondary active stabilizing gimbal 3-1 work together to counteract the low-frequency shaking and high-frequency jitter caused by the start and stop of the patrol platform base 1-1 and the movement of the segment assembly machine 9-3, so that the global array camera 3-2 can always obtain a clear motion image of the segment 9-1 to be installed. After the segment 9-1 to be installed is assembled in place, when performing fine joint inspection, the active dynamic motion decoupling and disturbance suppression algorithm enters the "high-precision locking mode". The controllers of the three motion units switch to higher gain, and the disturbance observer works at full capacity to "lock" the macro-micro fusion detection module 3 at the end above the target detection area. Its image stabilization accuracy can reach the sub-millimeter level, creating excellent measurement conditions for the 3D infrared scanner 3-3.

[0191] The working principle of the segment misalignment detection algorithm is as follows:

[0192] The segment misalignment detection algorithm is responsible for intelligently identifying, locating, and accurately quantifying various geometric deviations of segment joints from macroscopic images and microscopic viewpoints. It follows a process of "from macro to micro, identification before measurement." First, it quickly identifies and roughly locates the joints, guiding the macro-micro fusion detection module 3 into position; then, it performs fine scanning and measurement; finally, the results are fed back to the adaptive edge computing modules 2-3, forming a quality closed loop.

[0193] The segment misalignment detection algorithm is based on computer vision and deep learning methods. It uses a trained deep convolutional neural network to perform pixel-level semantic segmentation on panoramic images captured by a global array camera 3-2. The network outputs a mask of the same size as the input image, where each pixel is labeled as "segment joint," "segment body," or "background." This method can roughly identify the segment joints, i.e., the target detection area, under various lighting conditions and surface contamination, and output their accurate two-dimensional image contours.

[0194] After obtaining the two-dimensional image contour of the seam, the algorithm uses a pre-calibrated hand-eye matrix. (Describing the transformation relationship between the camera coordinate system and the robot arm's end effector coordinate system) and the robot's kinematic model, the two-dimensional pixel coordinates (u,v) are converted into a three-dimensional spatial target point (X,Y,Z) in the robot arm's base coordinate system. This three-dimensional spatial target point (X,Y,Z) is sent to the circular track patrol platform 1, driving the adaptive image-stabilized robot arm 2 to move towards the target detection area. During the movement, the segment misalignment detection algorithm continuously performs visual servoing, dynamically adjusting the motion trajectory according to the real-time deviation of the joint position in the image to ensure final accurate alignment.

[0195] After the macro-micro fusion detection module 3 is stably aligned with the joint, the 3D infrared scanner 3-3 acquires a dense point cloud of the joint area. Using an iterative nearest-point algorithm, the measured point cloud is registered with the point cloud of the designed standard segment model. The optimal rigid body transformation is solved to minimize the distance between the two point clouds, eliminating installation errors and establishing a measurement benchmark. On the registered point cloud, the segment misalignment detection algorithm performs precise geometric analysis: two planes are fitted on the segments on both sides of the joint using the least squares method. The misalignment can then be calculated as the distance difference between the two planes in the normal vector direction. By calculating the 3D Euclidean distance of the joint contour in the point cloud, the joint width is statistically obtained. By analyzing the residual between the fitted plane and the measured point cloud, the flatness of the segment surface is evaluated.

[0196] The algorithm flow for the segment misalignment detection algorithm is as follows:

[0197] a. Rapid macroscopic identification and positioning.

[0198] Specifically, deep learning-based semantic segmentation of seams identifies seams by using a convolutional network with an encoder-decoder structure:

[0199]

[0200] The loss function uses a combination of cross-entropy and Dice loss:

[0201]

[0202] Where y is the actual seam location, "This is the joint location predicted by the segment misalignment detection algorithm. The loss function considers both "whether each pixel prediction is correct" (cross-entropy term) and "how much the predicted joint area overlaps with the real area" (Dice loss term).

[0203] The transformation matrix is ​​obtained through hand-eye calibration. Convert the image coordinates to the robot arm's base coordinate system:

[0204]

[0205] in, This is the transformation matrix for normalized global array camera coordinates. It eliminates the influence of the global array camera's 3-2 intrinsic parameters, maps pixel coordinates to the normalized global array camera plane, and then lifts them to 3D space using depth d, achieving the transformation from pixel to global array camera coordinates; that is, pixel → global array camera coordinates: using Transform (u,v,d) into the global array camera coordinate system. The hand-eye matrix is ​​used to transform the global array camera coordinate system to the robot arm base coordinate system, i.e., global array camera coordinates → robot arm coordinates: using Convert global array camera coordinates to robotic arm coordinates (X) b ,Y b Z b ).

[0206] Using a pre-calibrated transformation matrix (Hand-eye matrix) accurately converts a pixel with coordinates (u,v) and depth d on an image to its three-dimensional coordinates (X,V) in the robot arm's base coordinate system. b ,Y b Z b This allows us to obtain the specific location of the corresponding seam position seen in the global array camera 3-2 in real space, and then control the adaptive image stabilizing robotic arm 2 to move towards that specific location in real space.

[0207] b. Visual servoing guidance:

[0208] Design image-based visual servo control:

[0209] Where s represents the image feature. These are the desired features.

[0210] The control law is:

[0211] Where e is the segment feature in the image, i.e., the deviation between the current position and the desired position. It is the pseudo-inverse of a matrix (interaction matrix) that reflects the relationship between the motion of the global array camera 3-2 and image changes. The above formula calculates the required speed v of the robotic arm's end effector 2-2 in real time based on the image deviation e. c This forms a closed-loop control system. This allows the detection system to automatically compensate for the instability of the segment assembly machine 9-3's movement, ensuring that the segment to be installed 9-1 is always centered in the field of view during movement.

[0212] c. Precise extraction of microscopic geometric features:

[0213] A robust ICP variant algorithm is used for point cloud registration and model matching:

[0214]

[0215] Wherein, weight w i Calculations are based on point-to-point distance and normal vector similarity. An iterative nearest-point algorithm is used to find a set of rotations R and translations t such that the scanned point cloud p... i With design model q c(i)Optimal matching. Precise extraction of micro-geometric features eliminates sensor installation errors and systematic positioning errors, establishing an accurate benchmark for all subsequent measurements.

[0216] Robust least squares was used for plane fitting on both sides of the joint.

[0217]

[0218] Where ρ is the Huber loss function.

[0219] Calculation of misalignment:

[0220]

[0221] On the registered point cloud, two planes are fitted using point cloud on both sides of the seam. and The misalignment δ is calculated by taking the average difference between the distances from all points to the opposite plane. This method makes full use of all three-dimensional data, rather than a single profile, thus obtaining a statistically more robust and accurate misalignment result with sub-millimeter precision.

[0222] d. Analysis of joint geometric characteristics:

[0223] Use standard deviation analysis to analyze width consistency:

[0224]

[0225] Among them, w i It is a single width value measured at different locations along the seam. σ is the arithmetic mean of the widths at all measurement points, i.e., the "average width." The core function of this formula is to assess the uniformity of the seam width. An ideal seam should have the same width everywhere. w The smaller the value of σ, the closer the width values ​​of all measurement points are to the average value, and the more uniform the joint width; conversely, the larger the value of σ, the more uniform the joint width. w The larger the value, the more inconsistent the width of the joint, indicating poor uniformity. If σ w If the threshold is exceeded, the segment misalignment detection algorithm will determine that the joint width is inconsistent. This may mean that the segment 9-1 to be installed has been twisted, locally squeezed, or improperly installed with the waterstop during installation. This provides an early warning to the detection system, indicating that attention needs to be paid to the long-term sealing performance of the joint.

[0226] e. Straightness error analysis:

[0227]

[0228] The segment misalignment detection algorithm uses the least squares method to analyze all 3D point cloud data points p on the joint centerline.i Fit an optimal three-dimensional straight line L fit Then, calculate the distance from each point on the center line to the fitted straight line L. fit vertical distance dist(p) i ,L fit Finally, the maximum value among all these distances is taken as the straightness error ε. straight This formula is used to directly measure whether a seam is straight, ε. straight The smaller the value, the closer the joint centerline is to a perfect straight line; the larger the value, the more obvious the bend or serpentine twist of the joint. This formula provides data feedback for the tunnel boring machine's deviation correction.

[0229] f. Coplanarity assessment:

[0230]

[0231] This formula is applicable to evaluating the degree of deviation of a set of points from an ideal plane.

[0232] First, the least squares method is used to fit an optimal plane to the point cloud data of one side of the joint surface. The mathematical expression of this plane is n·X + d = 0, where n is the unit normal vector of the plane and d is a constant term. For a point p belonging to this side surface... i Substituting the absolute value obtained into the plane equation |n·p i + d|, geometrically, represents the perpendicular distance from that point to the fitted plane. ε flat This refers to the maximum distance from all points to the plane, used to assess whether the tunnel inner wall formed after the 9-1 segment to be installed is smooth and flat. ε flat The smaller the value, the smoother the surface of the segment; the larger the value, the more likely there is a local bulge or depression.

[0233] Example 1: The present invention was applied to the segment assembly operation of a tunnel boring machine in a cross-river shield tunnel project with a diameter of 6.7 meters.

[0234] The ring-rail patrol platform 1 is made of high-strength, lightweight alloy precision casting. The inner diameter of the ring-rail base 1-3 is 6.8 meters. It is firmly installed on the H-beam 4 at the tail of the tunnel boring machine using M20 high-strength shock-absorbing bolts (shock-absorbing connectors 8) equipped with polyurethane elastic buffer pads. The bolt spacing is 250 mm, ensuring the stability and vibration resistance of the connection. The guide rail 1-4 is made of high-hardness bearing steel with a ground surface, exhibiting an extremely low coefficient of friction and high wear resistance. It precisely engages with the grooves of the four sets of opposing guide rollers 1-2 at the bottom of the patrol platform base 1-1, forming the core guiding and support mechanism for the stable operation of the patrol platform base 1-1. The drive system consists of a 1.8 kW servo motor and a precision planetary reducer. The drive gear meshes with the high-precision geared rail 1-6 fixed on the ring rail base 1-3, ultimately converting the rotation of the servo motor into a smooth and precise "cruising" motion of the entire cruise platform base 1-1 and its upper load along the circular guide rail 1-4 at a speed of 0.8 meters per second, providing a foundation for achieving blind-spot-free detection.

[0235] On the circular track patrol platform 1, the adaptive image-stabilized robotic arm 2 is mounted on the patrol platform base 1-1 via vibration-damping bolts 5, forming the intelligent execution unit of the detection system. The robotic arm body 2-2 is made of carbon fiber composite material with topology optimization design, featuring a 6-DOF serial configuration, a 2.2-meter reach, and a 15-kilogram load capacity. Each arm joint integrates a high-precision servo motor, harmonic reducer, and 23-bit absolute encoder, ensuring a repeatability accuracy of ±0.05 mm at the end effector. Its core control is handled by the adaptive edge computing module 2-3, powered by the NVIDIA Jetson AGX Orin platform. This module communicates in real-time with each joint of the robotic arm body 2-2 and the IMU sensor network distributed throughout the arm via Gigabit Ethernet, running active dynamic motion decoupling and disturbance suppression algorithms. It can actively observe and compensate for complex vibrations from the tunnel boring machine, providing a stable working environment for the macro-micro fusion detection module 3 at the end effector of the robotic arm body 2-2. The end effector interface gripper 2-1 is a precision dual-axis image stabilizing joint that integrates pitch and rotation degrees of freedom. It is responsible for the final micro-angle compensation and the quick assembly and disassembly of the macro-micro fusion detection module 3.

[0236] The macro-micro fusion detection module 3, serving as the sensing terminal directly executing the detection task, is connected to the end of the robotic arm body 2-2 via a quick-release structure. The module integrates a two-stage active stabilization gimbal 3-1, driven directly by a voice coil motor, possessing millisecond-level response capability. It is specifically designed to counteract residual high-frequency micro-vibrations after image stabilization by the adaptive image-stabilized robotic arm 2, thus forming a "master-slave collaborative" pixel-level image stabilization system together with the adaptive image-stabilized robotic arm 2. The core sensing components of the macro-micro fusion detection module 3 include a global array camera 3-2 composed of five 12-megapixel global shutter CMOS sensors. Its ultra-wide field of view, combined with adjustable LED illumination, enables rapid acquisition of clear macroscopic images over a large area; and a 3D infrared scanner 3-3 based on VCSEL structured light technology, capable of acquiring high-density 3D point clouds of seam surfaces with an accuracy of ±0.05 mm within a working distance of 0.5 to 3 meters. All acquired raw images and point cloud data are losslessly transmitted to the edge computing module for processing via a high-speed data interface.

[0237] When the segment assembly machine 9-3 assembles the segment 9-1 to be installed, the detection system automatically starts. The circular track patrol platform 1 first rapidly transports the adaptive image-stabilized robotic arm 2 and the macro-micro fusion detection module 3 to the predetermined observation area. Under the control of the active dynamic motion decoupling and disturbance suppression algorithm, the adaptive image-stabilized robotic arm 2 maintains stable tracking during the movement of the segment 9-1 to be installed. After the segment 9-1 to be installed is in place, the detection system completes a macroscopic scan of the entire ring of segments within 15 seconds, capturing image data through the global array camera 3-2 and transmitting it in real time to the adaptive edge computing module 2-3. The segment misalignment detection algorithm based on the DeepLabv3+ architecture running in the adaptive edge computing module 2-3 then performs pixel-level semantic segmentation on the image, accurately identifying the two-dimensional contours of all joints, and converting them into three-dimensional coordinates in the robotic arm base coordinate system through the hand-eye calibration matrix, guiding the detection system into the fine detection stage. At this point, the ring-track patrol platform 1 and the adaptive image-stabilized robotic arm 2 work together for precise positioning, while the secondary active stabilizing gimbal 3-1 is fully activated to suppress high-frequency vibrations. After stabilization and locking, the 3D infrared scanner 3-3 scans the target joint. The generated high-precision point cloud data is iteratively registered with the nearest point and robustly fitted with a plane by an algorithm. Finally, key quality parameters such as the number of error platforms, joint width, straightness error, and flatness are accurately calculated. The entire process can complete the inspection of all joints in a single ring within 3 minutes, which is more than 8 times more efficient than manual inspection. All data is archived in real time and a comprehensive quality assessment report is generated, providing a precise, efficient, and automated intelligent inspection solution for tunnel construction quality.

[0238] This invention designs a collaborative architecture consisting of a "circular track patrol platform 1 (large-range motion), an adaptive image-stabilized robotic arm 2 (medium-low frequency vibration suppression), and a two-stage active stabilization gimbal 3-1 (high-frequency micro-motion compensation)." This collaborative architecture, through mechanical decoupling and advanced dynamic motion decoupling and disturbance suppression algorithms, can actively counteract the strong vibration interference caused by tunnel boring machine construction, providing a pixel-level stable working platform for the macro-micro fusion detection module 3, fundamentally solving the industry problem of fuzzy and distorted detection data in harsh environments.

[0239] This invention employs a macro-micro fusion detection module 3, integrating a large field-of-view global array camera 3-2 and a high-precision 3D infrared scanner 3-3 onto a secondary active stabilizing gimbal 3-1. The process first uses macroscopic images acquired by the global array camera 3-2 to quickly identify and locate seams, then automatically guides the 3D infrared scanner 3-3 for precise measurement, achieving a seamless transition from "rapid scanning" to "precise measurement." This process significantly improves detection efficiency, ensuring comprehensive coverage without blind spots.

[0240] The segment misalignment detection algorithm built into this invention combines deep learning semantic segmentation with high-precision 3D point cloud processing. This algorithm not only automatically identifies joints but also intelligently analyzes and accurately calculates multiple key quality indicators such as the number of misalignments, joint width, straightness, and flatness. This enables intelligent and quantitative evaluation and real-time feedback of segment assembly quality, providing significant support for construction quality control.

[0241] In summary, this invention constructs an active image stabilization system that combines mechanical decoupling with algorithmic vibration suppression, and deeply integrates macroscopic visual perception with microscopic three-dimensional measurement. Through an intelligent detection system that integrates advanced dynamic control, deep learning, and three-dimensional vision algorithms, it achieves efficient, high-precision, and fully automated detection of segment assembly quality in harsh tunnel construction environments.

[0242] The above are merely preferred embodiments of the present invention and are not intended to limit the scope of protection of the invention. Therefore, any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A high-precision tunnel segment joint quality detection system based on a steady image mechanical arm, characterized in that: The application relates to a macro-micro fusion detection system for a shield tunneling machine, which comprises a circular track cruising platform (1), an adaptive image-stabilizing mechanical arm (2) and a macro-micro fusion detection module (3). The circular track cruising platform (1) comprises a cruising platform pedestal (1-1) and a guide track, the guide track is installed on an H-shaped beam (4) at the tail of the shield tunneling machine through a shock-absorbing connecting piece (8), the guide track has a circular ring structure and is located in the shield tunneling machine, the cruising platform pedestal (1-1) is arranged on the guide track and can move along the circumference of the guide track; one end of the adaptive image-stabilizing mechanical arm (2) is installed on the cruising platform pedestal (1-1), and the macro-micro fusion detection module (3) is installed on the other end of the adaptive image-stabilizing mechanical arm (2), so that the macro-micro fusion detection module (3) can move synchronously with the cruising platform pedestal (1-1) and is arranged to face the joint between a to-be-installed segment (9-1) and a segment (9-2) that has been installed. The guide track comprises rail-holding rollers (1-2), a circular track base (1-3), a guide slide rail (1-4), a cruising driving system (1-5) and a precision toothed rail (1-6); the guide slide rail (1-4) and the precision toothed rail (1-6) both have a circular ring structure and are concentrically stacked on the circular track base (1-3) and fixed, the circular track base (1-3) is installed on the H-shaped beam (4) at the tail of the shield tunneling machine through the shock-absorbing connecting piece (8); a pair of rail-holding rollers (1-2) are arranged below the cruising platform pedestal (1-1) at intervals, and the rail-holding rollers (1-2) are respectively matched with the inner ring and the outer ring of the guide slide rail (1-4) to roll and engage; the cruising driving system (1-5) is installed on the cruising platform pedestal (1-1), and the cruising driving system (1-5) is in meshing transmission connection with the precision toothed rail (1-6); the adaptive image-stabilizing mechanical arm (2) is installed on the cruising platform pedestal (1-1). The adaptive image-stabilizing mechanical arm (2) comprises an end effector interface clamp jaw (2-1), a mechanical arm body (2-2) and an adaptive edge computing module (2-3); the base of the mechanical arm body (2-2) is installed on the cruising platform pedestal (1-1) of the circular track cruising platform (1), the mechanical arm body (2-2) is composed of a plurality of arm joints driven by servo motors in series, and an inertial measurement sensor is embedded in each joint; the adaptive edge computing module (2-3) is arranged on and connected with the mechanical arm body (2-2), a segment misjoint detection algorithm and an active dynamic motion decoupling and disturbance suppression algorithm are integrated in the adaptive edge computing module (2-3); the end effector interface clamp jaw (2-1) is installed at the end of the mechanical arm body (2-2) away from the cruising platform pedestal (1-1), a pitch joint and a rotation joint are integrated in the end effector interface clamp jaw (2-1), and the macro-micro fusion detection module (3) is installed at the end of the mechanical arm body (2-2) through the pitch joint and the rotation joint. The macro-micro fusion detection module (3) comprises a secondary active stabilizing holder (3-1), a global array camera (3-2) and a three-dimensional infrared scanner (3-3); the secondary active stabilizing holder (3-1) is detachably connected with the end effector interface clamp jaw (2-1), the global array camera (3-2) and the three-dimensional infrared scanner (3-3) are mounted on the secondary active stabilizing holder (3-1), and the global array camera (3-2) and the three-dimensional infrared scanner (3-3) are arranged to face the joint between the to-be-installed segment (9-1) and the installed segment (9-2).

2. A detection method using the tunnel segment joint quality high-precision detection system based on the steady image mechanical arm according to claim 1, characterized in that: The method comprises the following steps: Step 1: the to-be-installed segment (9-1) is positioned, that is, the to-be-installed segment (9-1) is transported to the assembling position; Step 2: the detection system receives a detection instruction, the ring rail cruising platform (1) drives the adaptive image stabilization mechanical arm (2) to move, and the positioning process of the to-be-installed segment (9-1) is monitored; Step 3: the to-be-installed segment (9-1) is assembled by the segment assembling machine (9-3); Step 4: the ring rail cruising platform (1) drives the adaptive image stabilization mechanical arm (2) to move along the guide rail (1-4), and the adaptive image stabilization mechanical arm (2) controls the macro-micro fusion detection module (3) at the tail end to move to the target detection area; Step 5: the macro-micro fusion detection module (3) performs preliminary detection on the target detection area of the joint of the to-be-installed segment (9-1); Step 6: based on the preliminary detection result, accurate positioning and image stabilization locking are performed by the adaptive image stabilization mechanical arm (2) and the macro-micro fusion detection module (3); Step 7: the macro-micro fusion detection module (3) performs microscopic measurement and data acquisition on the target detection area of the joint; Step 8: the to-be-installed segments (9-1) of the whole ring are detected in sequence according to the segment assembling sequence, after the joint detection of one ring segment is completed, steps 1-7 are repeated, and a quality detection report of each joint is generated in real time; Step 9: after the detection of all to-be-installed segments (9-1) is completed, the adaptive image stabilization mechanical arm (2) contracts all joints to a transportation posture, and the cruising platform base (1-1) returns to the initial position along the guide rail (1-4).

3. The method of claim 2, wherein: The step 2 comprises the following sub-steps: Step 21: when the to-be-installed segment (9-1) is transported to the assembling position, the ring rail cruising platform (1) drives the adaptive image stabilization mechanical arm (2) and the macro-micro fusion detection module (3) to move to the predetermined observation position under the control of the adaptive edge computing module (2-3); Step 22: the adaptive image stabilization mechanical arm (2) keeps stable tracking in the movement process of the to-be-installed segment (9-1) through active dynamic motion decoupling and disturbance suppression algorithm; Step 23: the global array camera (3-2) of the macro-micro fusion detection module (3) continuously photographs the to-be-installed segment (9-1) in motion under the image stabilization guarantee of the secondary active stabilizing holder (3-1), and transmits real-time image data stream to the adaptive edge computing module (2-3), thereby realizing the monitoring of the positioning process of the to-be-installed segment (9-1).

4. The method of claim 2, wherein: In step 5, the macro-micro fusion detection module (3) performs rapid macroscopic scanning on the circumferential and tangential seams of the pipe segment (9-1) to be installed: the global array camera (3-2) acquires panoramic images and transmits the panoramic image data to the adaptive edge computing module (2-3) in real time for preliminary identification of the seams, and records the macroscopic position and attitude data of the pipe segment (9-1) to be installed; In step 6, based on the preliminary detection results in step 5, i.e., the macroscopic position and attitude data of the pipe segment (9-1) to be installed, the fine detection stage is entered: the adaptive edge computing module (2-3) calculates the accurate three-dimensional coordinates of the seams through the hand-eye calibration matrix, the ring track cruising platform (1) performs coarse positioning, moves the adaptive image stabilization robot arm (2) to the target area, the robot arm body (2-2) of the adaptive image stabilization robot arm (2) performs fine pose adjustment through the movement of each arm joint, and the pitch joint and rotation joint built-in the end effector interface gripper (2-1) perform micro-angle compensation, the secondary active stabilization gimbal (3-1) activates high-frequency vibration suppression, and pixel-level stabilization is achieved.

5. The method of claim 2, wherein: In step 7, the adaptive edge computing module (2-3) sends measurement instructions to the macro-micro fusion detection module (3), the three-dimensional infrared scanner (3-3) performs high-density point cloud collection on the target seam, and transmits all original three-dimensional point cloud data to the adaptive edge computing module (2-3) without loss, the pipe segment misalignment detection algorithm processes data in real time to calculate the misalignment amount and seam width; In step 8, after completing the seam detection of one ring, the entire detection system moves to the next ring pipe segment assembly position along with the shield machine, and automatically plans the seam detection path of the next ring pipe segment, and the cruising platform base (1-1) moves accurately to the next detection position along the guide rail (1-4).

6. The method of claim 3, wherein: The algorithm flow of the active dynamic motion decoupling and disturbance suppression algorithm is as follows: a. Establish the dynamic model of the entire detection system: first, establish the dynamic equation including mass, damping and stiffness for the entire ring track cruising platform (1), adaptive image stabilization robot arm (2) and secondary active stabilization gimbal (3-1): ； where q is the generalized coordinate representing the position of each joint; M(q) is the system inertia matrix; is the Coriolis and centrifugal force matrix; G(q) is the gravity term, τ is the control torque, is the external disturbance; b. Dynamic motion decoupling: the control law is designed as follows: ； Where u is the new control input; Substitute the above control law into the original dynamic equation to obtain the decoupled linear system: ； The edge computing module controls the drive system to actively output a torque τ; Further design u to track the desired trajectory and suppress disturbances; c. Disturbance observer design: Define the estimation value of the disturbance observer as with the update law: ； It estimates the magnitude and direction of the external vibration disturbance in real time by comparing the difference between the actual position q of the adaptive steady camera robot (2) and the position it should have been in an ideal undisturbed environment ​​​ where L is the gain matrix of the disturbance observer, is the velocity estimate calculated by the dynamic model; a disturbance observer structure is adopted as follows: ； where K o is the correction gain of the disturbance observer; d. Feedforward-feedback compound control, using feedforward-feedback control structure; feedback control uses PD control, and feedforward control uses the output of disturbance observer; the control input u is designed as: ; where K p and K d are control parameters; q̈ d , q̇ d , q d are target trajectories; Introduce an adaptive notch filter algorithm, and the transfer function of the notch filter algorithm is: ； wherein is a notch frequency, being the frequency of the vibration to be suppressed; and is a damping coefficient; When the pipe segment assembling machine (9-3) moves: the active dynamic motion decoupling and disturbance suppression algorithm estimates the torque disturbance caused by the movement of the pipe segment assembling machine (9-3) through the disturbance observer, and offsets it through feedforward compensation, while the notch filter suppresses the main vibration frequency to ensure the stability of the end of the robot arm body (2-2); When the shield machine advances: the low-frequency vibration generated by the advance is suppressed by feedback control and feedforward compensation, and the high-frequency vibration is handled by the secondary active stabilization gimbal (3-1). The active dynamic motion decoupling and disturbance suppression algorithm continuously receives acceleration and angular velocity data from a network of IMU sensors deployed in the joints of the base, body, and end of the robotic arm body (2-2); through optimal fusion by a Kalman filter, the amplitude, frequency, and phase of the vibration are accurately estimated; based on this, a feedforward compensation signal is generated and directly superimposed into the current loop of the servo motor of each joint; When the segment erector (9-3) moves the to-be-installed segment (9-1): the active dynamic motion decoupling and disturbance suppression algorithm controls the loop track platform (1) to perform smooth tracking motion, while the adaptive image stabilization robotic arm (2) and the secondary active stabilization gimbal (3-1) work together to offset the low-frequency shaking and high-frequency jitter caused by the start and stop of the loop track platform (1-1) and the movement of the segment erector (9-3), so that the global array camera (3-2) can always obtain clear motion images of the to-be-installed segment (9-1).

7. The method of claim 5, wherein: The algorithm flow of the segment misalignment detection algorithm is: a. Macroscopic rapid identification and positioning: Joint semantic segmentation based on deep learning, which identifies joints by using an encoder-decoder structure convolutional network: ； The loss function uses a combination of cross-entropy and Dice loss: ； where y is the true joint position, is the joint position predicted by the joint misalignment detection algorithm. Obtain transformation matrix ᵇT by hand-eye calibration c Convert image coordinates to robot base coordinates: ； Utilize the pre-precise calibration of the transformation matrix ᵇT c , a pixel point with image coordinates , depth d is accurately converted to the three-dimensional coordinates of the robot base coordinate system , so as to obtain the specific position of the corresponding seam position in the real space seen in the global array camera (3-2), and then control the adaptive image stabilization robot (2) to move to the specific position in the real space; b. Visual servoing guidance: Designing image-based visual servoing control: ; wherein s is an image feature, is a desired feature; The control law is: ; where e is the image feature of the tube sheet, i.e. the deviation of the current position from the desired position, is the pseudo-inverse of a matrix reflecting the relationship between the global array camera (3-2) motion and the image change; c. Microscopic geometric feature accurate extraction: Robust ICP variant algorithm is used for point cloud registration and model matching: ； wherein the weights Based on the distance between the point pair and the normal vector similarity calculation; through the iterative closest point algorithm, find a set of rotation R and translation t, so that the point cloud obtained by scanning With the design model Best match; Robust least squares is used for fitting the planes on both sides of the joint: ； Where ρ is the Huber loss function; Misalignment calculation: ； Two planes are fitted to the point cloud on both sides of the joint and The amount of misalignment δ is the average of the distances of all points to the opposite plane. d. Joint geometric property analysis: Standard deviation is used to analyze the width consistency: ； wherein, is the single width value measured at different locations of the joint; is the arithmetic mean of all measured width values, i.e. the "average width; The smaller the value of is, the closer the width values of all measured points are to the average value, and the more uniform the joint width is. Conversely, The larger the value of is, the more the joint width varies, and the less uniform it is. If exceeds the threshold value, the pipe joint misalignment detection algorithm determines that the joint width is not uniform. e. Straightness error analysis: ； The segment misalignment detection algorithm uses a least squares method to fit a straight line in three-dimensional space to all the three-dimensional point cloud data points p i that lie on the centerline of the joint ; then, the perpendicular distance of each point on the centerline to the fitted straight line is calculated ; finally, the maximum value of all these distances is taken as the straightness error , The smaller the value, the closer the joint centerline is to a perfect straight line; the larger the value, the more obvious the bending or serpentine twisting of the joint. f. Coplanarity evaluation: ； First, the point cloud data of the surface of one side of the jointed segment is fitted to a best plane using the least square method; the mathematical expression of the plane is where n is the unit normal vector of the plane and d is a constant term; for a point p belonging to the surface of the side i , the absolute value of the point substituted into the plane equation is , which is the perpendicular distance from the point to the fitted plane in geometry; that is, the maximum value of the distances from all points to the plane, is used to evaluate whether the inner wall of the tunnel formed after the installation of the segment (9-1) is smooth and flat; The smaller the value is, the smoother the surface of the segment at the position is; the larger the value is, the more local bulges or depressions exist.

Images