Method for inspecting the inner wall of small diameter pipes of a bionic robotic snake based on a vision system

Abstract

The application relates to the field of special robot technology, in particular to a small-pipe-diameter pipeline inner wall detection method of a bionic mechanical snake based on a vision system, which comprises the following steps: one, designing a bionic mechanical snake structure; two, designing an intelligent gait planning algorithm to perform multi-modal adaptive motion control; three, designing a multi-modal detection system, the system fusing vision, thermal imaging and inertial navigation; four, based on the bionic mechanical snake, the intelligent gait planning algorithm and the multi-modal detection system, detecting the inner wall complex working conditions of the small-pipe-diameter pipeline. The application can preferably detect the inner wall of the small-pipe-diameter pipeline.

Description

Technical Field

[0001] This invention relates to the field of special robot technology, and more specifically, to a method for detecting the inner wall of a small-diameter pipe using a vision-based bionic robotic snake. Background Technology

[0002] In the field of small-diameter pipe inner wall inspection, existing technologies mainly rely on wheeled / tracked inspection robots and traditional bionic robotic snakes, but both types of solutions have significant drawbacks. Wheeled / tracked robots rely on a rigid chassis and fixed drive wheels. Although they can achieve rapid linear movement (5-8 cm / s), in narrow pipe diameters of 50-100 mm or bends with a curvature radius <1.5 times the pipe diameter, the probability of jamming exceeds 25% due to the mismatch between the wheelbase and the pipe diameter. Their sensor viewing angle is fixed, covering only 60%-70% of the pipe inner wall area, and only supports surface defect detection. The missed detection rate for microcracks or latent thermal faults smaller than 0.5 mm is as high as 20%, making it difficult to adapt to complex pipe network environments. Traditional bionic robotic snakes, while possessing flexible joint structures, have limited parallel joint amplitude (≤±70°), a single gait (supporting only fixed meandering motion), a slippage frequency of up to 35% on smooth inner walls, and lack of driven wheels, resulting in a failure rate exceeding 30% when navigating narrow sections of 50-80mm. In terms of detection, they rely on a monocular camera, which can only identify visible cracks larger than 0.8mm, lacking thermal imaging or ultrasonic detection capabilities, and have a positioning error of >10cm in a 100m pipeline, failing to meet the requirements for accurate detection. Summary of the Invention

[0003] The present invention provides a method for detecting the inner wall of small-diameter pipes using a bionic robotic snake based on a vision system, which can better detect the inner wall of small-diameter pipes.

[0004] According to the present invention, a method for detecting the inner wall of a small-diameter pipe using a vision-based bionic robotic snake includes the following steps:

[0005] I. Design of a biomimetic robotic snake structure;

[0006] II. Design an intelligent gait planning algorithm for multimodal adaptive motion control;

[0007] III. Design a multimodal detection system that integrates vision, thermal imaging, and inertial navigation;

[0008] IV. Based on biomimetic robotic snake, intelligent gait planning algorithm and multimodal detection system, the complex working conditions of the inner wall of small diameter pipeline are detected.

[0009] Preferably, in step one, the bionic robotic snake structure includes a head joint, multiple middle joints, and a tail joint. The middle joints are orthogonally connected and support a rotation range of ±85°.

[0010] As a preferred option, the head joint features a layered cabin design, including a control cabin, an optical coupler isolation cabin, and a sensor cabin, with an IP68 waterproof outer shell connected by bolts.

[0011] The middle joint includes a horizontal joint with a driven wheel and a creeping joint without wheels. The horizontal joint with a driven wheel is driven by a dual-axis servo motor. The hollow design in the middle of the joint accommodates the servo motor and has reserved holes for wire insertion. The outer shell, driven wheel and main joint are assembled by round-head bolts and lock nuts. The creeping joint without wheels moves in the vertical direction and is orthogonally connected to the horizontal joint through a servo disk to simulate the arching action of an inchworm.

[0012] The tail joint and battery compartment have reserved space for buoyancy adjustment and are connected to the rudder of the middle joint via a linkage.

[0013] As a preferred option, in step two, the intelligent gait planning algorithm employs a composite gait strategy:

[0014] During meandering motion, based on the Hirose serpentine curve curvature equation:

[0015]

[0016] s is the arc length parameter, used to determine the curve position; L is the length of the robot snake, which is involved in the curvature quantization calculation;

[0017] Dynamically adjust the number of waveforms K n With an initial swing angle α0, to achieve a minimum curvature radius of 0.8 times the pipe diameter for cornering, the servo angle function is:

[0018]

[0019] θ i t represents the angle value of the i-th servo motor; t represents time; and i represents the joint number.

[0020] During the peristaltic motion, the three wheelless joints arch upwards through a sequence of sinusoidal functions:

[0021]

[0022] In the gait switching mechanism, inertial navigation monitors the pipe curvature and diameter in real time and automatically switches the motion mode.

[0023] In friction adaptive control, the roughness of the inner wall is analyzed by visual image texture analysis, and the pressure of the driven wheel and the joint swing amplitude are dynamically adjusted.

[0024] Preferably, in step three, the multimodal detection system includes a vision module, a thermal imaging module, and an inertial navigation module.

[0025] The vision module uses a high-definition fisheye camera and is equipped with the YOLOv5 algorithm to identify cracks in real time.

[0026] The thermal imaging module uses infrared sensors to detect abnormal temperatures, accurately locate leaks, and solve the problem of missing hidden faults.

[0027] The inertial navigation module uses a nine-axis IMU to monitor the snake's position and tilt angle in real time, providing reference data for gait planning and defect localization.

[0028] The multimodal detection system uses a joint detection rule, which employs both visual texture feature extraction and thermal imaging temperature gradient analysis for dual judgment. This is combined with an extended Kalman filter to fuse multi-source data. The state equation and observation equation of the extended Kalman filter are as follows:

[0029]

[0030] x k+1 f(x) represents the state vector of the system at time k+1. k ) is the state transition function; w k z is the process noise vector; k h(x) is the observation vector at time k; k ) is the observation function; v k This is the observed noise vector.

[0031] The beneficial effects of this invention are as follows:

[0032] This invention achieves a systematic breakthrough in addressing the technical bottlenecks of existing small-diameter pipeline inspection through the deep integration of biomimetic structural innovation, intelligent gait planning, and multimodal detection technology. The specific effects are as follows:

[0033] I. Significantly improved efficiency in handling complex pipeline networks

[0034] The robot's mobility and maneuverability are enhanced by employing lightweight orthogonal joints with a ±85° rotation range, combined with a "winding-creeping" composite gait planning algorithm, enabling it to adapt to complex pipe networks with diameters ranging from 50 to 300 mm. In small-radius bends (curvature radius ≥ 0.8 times the pipe diameter), by dynamically adjusting joint amplitude (maximum amplitude 50°) and gait parameters, the bend attitude error is < 5°, resulting in a 40% improvement in maneuverability compared to traditional wheeled robots (minimum bend radius 1.5 times the pipe diameter).

[0035] The meandering movement speed reaches 10.14cm / s (servo motor rotation frequency 0.314Hz, swing amplitude 40°), which is suitable for rapid inspection in open pipes; the creeping gait propels the snake at a speed of 3.5cm / s in a 100mm narrow straight pipe, which is 1.9 times faster than traditional bionic robotic snakes and solves the problem of getting stuck in narrow spaces.

[0036] II. Multimodal detection accuracy achieves a breakthrough of orders of magnitude

[0037] The snake-head integrated vision-thermal imaging-inertial navigation composite sensor group, which enables defect identification and localization, constructs a joint judgment system of "visual texture analysis + thermal imaging gradient detection":

[0038] The high-definition camera (12 megapixels, 1920×1080 resolution) combined with the YOLOv5 algorithm can identify cracks ≥0.2mm (mAP@0.5=96%), which is 4 times more accurate than traditional monocular vision (recognition threshold 0.8mm), and the false negative rate is reduced from >10% to below 3%.

[0039] The ±1.5℃ thermal imager can detect temperature anomalies with ΔT≥3℃. Combined with the extended Kalman filter algorithm to fuse multi-source data, the leak point location accuracy is ≤2cm, which is 5 times higher than the traditional single-modal detection (error>10cm).

[0040] Multi-gait collaborative inspection efficiency is achieved by real-time attitude monitoring (accuracy ±0.5°) via an inertial navigation unit (IMU). The robotic snake can automatically switch gaits according to the pipe curvature: it adopts a meandering motion in straight pipes (inspection speed 10.14 cm / s) and switches to a creeping gait (inspection speed 3.5 cm / s) in bends or narrow sections. The inspection time for a single 100m pipe section is reduced from more than 8 hours to less than 5 hours, improving efficiency by 62.5%.

[0041] III. Intelligent Control and Algorithm Optimization to Improve Reliability

[0042] The motion control algorithm's advantage lies in its use of the Hirose serpentine curve curvature equation, combined with pipe parameters (pipe diameter D, radius of curvature R) to dynamically adjust the number of waveforms K. n With the initial swing angle α0, adaptive optimization of gait parameters is achieved, ensuring that the motion stability error is less than 3% under different pipe diameters.

[0043] The creeping gait, through the angle function of three wheelless joints, allows for a climbing speed of 0.2m / s within a 150mm diameter pipe. When crossing obstacles, it increases the swing amplitude (to 50°) to ensure no rollover and adapts to complex terrain with a slope of ≤20°.

[0044] The data fusion and control precision control system uses an STM32 development board and an ESP8266 module to achieve precise control of 8 servos (PWM signal accuracy ±1%) through an asynchronous serial bus. Combined with an extended Kalman filter algorithm, the joint angle control error is controlled within ±2°, ensuring the spatiotemporal synchronization of gait planning and sensor detection.

[0045] IV. Significantly Enhanced Environmental Adaptability and Engineering Applicability

[0046] The snake-head is designed for stable operation under complex working conditions. Its IP68 waterproof hull (withstanding a water depth of 1m) and tail hull buoyancy adjustment feature support both dry and wet pipelines. The battery compartment provides ≥4 hours of battery life and can inspect 200m of pipeline in a single operation, which is 30% longer than traditional equipment.

[0047] On inner walls made of different materials such as concrete and plastic, the gait amplitude is automatically adjusted (adjustment range ±10°) through a sensor linkage strategy to ensure the friction coefficient is matched, thereby improving movement stability by 70%.

[0048] The modular design and expandable snake-body joints adopt standardized interfaces (round head bolt fastening), supporting quick replacement of driven wheel modules and ultrasonic thickness measurement modules to meet the needs of different inspection scenarios; the AI ​​image recognition system automatically marks defects and generates heat maps, reducing manual intervention and promoting the automation of the inspection process. Attached Figure Description

[0049] Figure 1 This is a flowchart illustrating a method for detecting the inner wall of a small-diameter pipe using a vision-based bionic robotic snake, as described in this embodiment.

[0050] Figure 2 This is a schematic diagram of the biomimetic robotic snake structure in the embodiment. Detailed Implementation

[0051] To further understand the content of this invention, a detailed description of the invention will be provided in conjunction with the accompanying drawings and embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0052] Example

[0053] like Figure 1 As shown, this embodiment provides a method for detecting the inner wall of a small-diameter pipe using a vision-based bionic robotic snake, which includes the following steps:

[0054] I. Design of a biomimetic robotic snake structure;

[0055] like Figure 2 As shown, the biomimetic robotic snake structure includes a head joint 1, multiple middle joints 2, and a tail joint 3. The middle joints 2 are orthogonally connected, supporting a rotation range of ±85°. The head joint 1 has a layered cabin design, including a control cabin, an optical coupler isolation cabin, and a sensor cabin. The IP68 waterproof shell is connected by bolts.

[0056] The middle joint 2 includes a horizontal joint with a driven wheel and a creeping joint without wheels. The horizontal joint with a driven wheel is driven by a dual-axis servo motor. The hollow design in the middle of the joint accommodates the servo motor and has reserved wire insertion holes. The outer shell, driven wheel and main joint are assembled by round head bolts and lock nuts. The creeping joint without wheels moves in the vertical direction and is orthogonally connected to the horizontal joint through a servo disk to simulate the arching action of an inchworm.

[0057] The tail joint 3 has a battery compartment with reserved buoyancy adjustment space, and is connected to the rudder of the middle joint via a linkage.

[0058] The biomimetic robotic snake structure employs a lightweight orthogonal joint design, using a modular combination of a horizontal joint with driven wheels (±85° rotation range) and a wheelless peristaltic joint to construct a detachable and modular snake body. The joint with driven wheels is driven by a dual-axis servo motor. A hollow design in the middle of the joint accommodates the servo motor and provides pre-drilled wiring holes to ensure circuit stability. The driven wheel housing is secured to the main joint with round-head bolts, allowing for quick disassembly during maintenance, improving maintenance efficiency by 70% compared to traditional integrated joints. The wheelless peristaltic joint handles vertical movement, enabling inchworm-like propulsion. Combined with a 3D-printed weather-resistant resin shell (water absorption <0.5%) and anti-slip texture, it improves movement stability by 40% in scaled and humid environments. The snake's head features a layered design with a control compartment, a sensor compartment, and an optocoupler isolation compartment, supporting sensor module replacement within 3 minutes. The tail compartment serves as a battery compartment and provides space for buoyancy adjustment, enabling universal use in both dry and wet pipelines.

[0059] II. Design an intelligent gait planning algorithm for multimodal adaptive motion control;

[0060] In step two, the intelligent gait planning algorithm employs a composite gait strategy:

[0061] During meandering motion, based on the Hirose serpentine curve curvature equation:

[0062]

[0063] s is the arc length parameter, used to determine the curve position; L is the length of the robot snake, which is involved in the curvature quantization calculation;

[0064] Dynamically adjust the number of waveforms K n With an initial swing angle α0, a minimum curvature radius of 0.8 times the pipe diameter is achieved for cornering (compared to 1.5 times in existing technology). The joint swing amplitude adaptively adjusts during cornering to ensure unobstructed passage. The servo angle function is:

[0065]

[0066] θ i t is the angle value of the i-th servo motor (unit: degrees); t is the time (unit: seconds); i is the joint number, which distinguishes different servos. In meandering motion, it corresponds to horizontal joints 1, 3, 5, and 7, and in creeping motion, it corresponds to wheelless joints 2, 4, and 6.

[0067] During the peristaltic motion, the three wheelless joints arch upwards through a sequence of sinusoidal functions:

[0068]

[0069] The propulsion speed reached 3.5 cm / s in a narrow 50 mm straight tube, which is 1.9 times faster than the traditional peristaltic gait. At the same time, by analyzing the roughness of the inner wall through visual image texture analysis, the driven wheel pressure and joint amplitude were dynamically adjusted to reduce the slippage frequency on the smooth inner wall from 35% to 12%.

[0070] In the gait switching mechanism, inertial navigation monitors the pipe curvature and diameter in real time and automatically switches the motion mode.

[0071] In friction adaptive control, the roughness of the inner wall is analyzed by visual image texture analysis, and the pressure of the driven wheel and the joint swing amplitude are dynamically adjusted.

[0072] III. Design a multimodal detection system that integrates vision, thermal imaging, and inertial navigation;

[0073] In step three, the multimodal detection system includes a vision module, a thermal imaging module, and an inertial navigation module;

[0074] The vision module uses a high-definition fisheye camera and is equipped with the YOLOv5 algorithm to identify cracks in real time.

[0075] The thermal imaging module uses infrared sensors to detect abnormal temperatures, accurately locate leaks, and solve the problem of missing hidden faults.

[0076] The inertial navigation module uses a nine-axis IMU to monitor the snake's position and tilt angle in real time, providing reference data for gait planning and defect localization.

[0077] The multimodal detection system uses a joint detection rule, which employs both visual texture feature extraction and thermal imaging temperature gradient analysis for dual judgment. This is combined with an extended Kalman filter to fuse multi-source data. The state equation and observation equation of the extended Kalman filter are as follows:

[0078]

[0079] x k+1 This represents the state vector of the system at time k+1. It contains various state information of the system at that time, such as position and velocity (depending on the system modeling). It is the state of the next time step predicted based on the state at time k.

[0080] f(x k (x) is the state transition function, which describes the system's state x from time k. k How the system evolves to the state at time k+1 reflects its dynamic characteristics and is determined based on the system's physical laws or mathematical models.

[0081] w kLet z be the process noise vector, representing the uncertainties and random disturbances introduced during state transitions, such as sensor errors and environmental interference. It is typically assumed to follow a certain probability distribution (e.g., a Gaussian distribution). k The observation vector at time k is the data actually measured by sensors and other devices, which reflects some information about the system state, but often contains noise;

[0082] h(x k (x) is the observation function, which describes the system state x. k With the observed value z k The mapping relationship between them is used to predict observations based on the system state;

[0083] v k The observation noise vector represents the noise introduced during the observation process. Due to factors such as sensor accuracy, there is a deviation between the observed value and the true value. It is generally assumed that the observed value follows a certain probability distribution (such as Gaussian distribution).

[0084] The 12-megapixel high-definition camera (supporting 360° rotation) at the front of the snake head works in conjunction with a ±1.5℃ thermal imager, combining the YOLOv5 algorithm to achieve high-precision identification of cracks ≥0.2mm (mAP@0.5=96%). The thermal imaging module detects temperature anomalies ΔT≥3℃ and locates leak points with an accuracy ≤3cm, solving the problem of missed detection by a single sensor. The nine-axis IMU monitors attitude and position in real time, and establishes a state equation x by fusing multi-source data through an extended Kalman filter algorithm. k+1 =f(x) k )+w k With the observation equation z k =h(x k )+v k It achieves centimeter-level positioning (error ≤ 2cm), improving accuracy by 5 times compared to traditional single-modal detection. Furthermore, the sensor-linked control strategy automatically adjusts the gait amplitude based on the pipe material (concrete, plastic, etc.), improving stability by 70% in different internal wall environments. A single charge provides ≥ 4 hours of runtime and can cover 200m of pipe inspection.

[0085] IV. Based on biomimetic robotic snake, intelligent gait planning algorithm and multimodal detection system, the complex working conditions of the inner wall of small diameter pipeline are detected.

[0086] 4.1. Water Entry and Admission Phase (0-8 minutes)

[0087] Connection and debugging: The robotic snake is inserted into the pipe through the inspection port, the ballast block of the tail compartment is adjusted to neutral buoyancy (contact force 5N), and the snake head sensor group performs a self-test (camera white balance calibration, thermal imager temperature calibration).

[0088] Motion control: A meandering gait (amplitude 35°, frequency 0.2Hz) is used to pass through the gate connection, with the joint angle function as follows:

[0089]

[0090] Passing through a 150mm narrow section without collision within 2 minutes, with the IMU monitoring the attitude deviation in real time as <3°, verifying the passability of the orthogonal joint within a ±85° rotation range.

[0091] 4.2 Complex Environment Testing Phase (8-40 minutes)

[0092] (1) Straight pipe section with moss (100m)

[0093] Gait switching: Upon detection of stable tube diameter (200mm) and inner wall roughness Ra > 5μm, the gait switches to a creeping gait, with the three wheelless joints arching at a 40° amplitude and a 0.25Hz frequency. Joint angle function:

[0094]

[0095] The driven wheel adheres to the inner wall with a pressure of 5N, and propels the vehicle at a speed of 3.5cm / s, which is 1.9 times faster than the traditional creeping gait, with a slippage frequency of less than 10%.

[0096] Multimodal detection:

[0097] The high-definition camera captures images at 10 frames per second, and the YOLOv5 model identifies cracks ≥0.2mm in real time and marks two 0.3mm microcracks.

[0098] The thermal imager scanned the temperature field at 2Hz and found three abnormal areas with ΔT = -2℃ to -3℃ (characteristics of seepage points).

[0099] (2) 60° elbow section (curvature radius 240mm = 1.2 times pipe diameter)

[0100] Curvature adaptive control: Inertial navigation triggers a meandering turning algorithm, calling the curvature equation. Where K n =4, α0=50°, joint swing is dynamically adjusted to 75°, cornering is completed within 15 seconds, posture error <5°, no jamming.

[0101] Initial defect screening: The thermal imager detected an abnormal temperature gradient (ΔT = 3.5℃) on the inside of the elbow, which was marked as a suspected leakage area, triggering the detailed inspection process.

[0102] 4.3. Detailed inspection of suspected defects (40-45 minutes)

[0103] 360° rotating scan of the snake head: The snake head servo is controlled to rotate at 5° / second, and a high-definition camera performs macro photography on the suspected area (resolution increased to 2592×1944). Combined with the edge detection algorithm, a 0.4mm crack is confirmed. The thermal imager scans simultaneously and confirms that the temperature in the crack area is 2.5℃ lower than the surrounding area, which is consistent with the leakage characteristics.

[0104] Multi-source data fusion: Extended Kalman filter fuses IMU location data (error ≤ 2cm), visual odometry and pipeline BIM model to accurately locate defect location (error 3cm) and generate three-dimensional coordinates (X = 12.35m, Y = 0.8m, Z = 1.2m).

[0105] 4.4 Data Processing and Evacuation Phase (45-50 minutes)

[0106] Data processing: The detection data is transmitted to the monitoring host via the underwater acoustic communication module. The AI ​​system automatically generates a defect heat map, marking 5 cracks (0.2-0.5mm) and corrosion areas (≥10mm). 2 There were 3 leaks and 2 seepage points, with a failure rate of 1.2% (only one 0.15mm microcrack was not detected).

[0107] Evacuation control: The tail compartment was weighted by 200g and switched to a reverse crawling gait. It returned to the maintenance port within 5 minutes. There was no scale buildup or blockage in the joints, which verified the anti-fouling ability of the modular structure.

[0108] Comparison of core performance indicators: as shown in the table below.

[0109]

[0110] This embodiment systematically solves the core pain points of existing equipment in small-diameter pipeline inspection through collaborative innovation in mechanical structure, motion control, and detection technology. It achieves a technological leap from "single detection" to "intelligent full inspection" and from "rigid motion" to "flexible adaptation," providing an efficient, accurate, and reliable intelligent solution for the safe operation of industrial pipelines.

[0111] The present invention and its embodiments have been described above illustratively. This description is not restrictive, and the figures shown are only one embodiment of the present invention; the actual structure is not limited thereto. Therefore, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the present invention, such designs should fall within the protection scope of the present invention.

Claims

1. A method for inspecting the inner wall of a small diameter pipe of a biomimetic robotic snake based on a vision system, characterized by: Includes the following steps: I. Design of a biomimetic robotic snake structure; The biomimetic robotic snake structure includes a head joint, multiple middle joints, and a tail joint. The middle joints adopt an orthogonal connection method, supporting a rotation range of ±85°. The head joint has a layered cabin design, including a control cabin, an optocoupler isolation cabin, and a sensor cabin. The IP68 waterproof shell is connected by bolts. The middle joints include a horizontal joint with driven wheels and a peristaltic joint without wheels. The horizontal joint with driven wheels is driven by a dual-axis servo motor. The hollow design in the middle of the joint accommodates the servo motor and provides a reserved wire insertion hole. The shell, driven wheels, and main joint are assembled by round-head bolts and lock nuts. The peristaltic joint without wheels moves in the vertical direction and is orthogonally connected to the horizontal joint through a rudder disk, simulating the arching action of an inchworm. The tail joint has a battery compartment with reserved buoyancy adjustment space and is connected to the rudder disk of the middle joint through a connecting rod. II. Design an intelligent gait planning algorithm for multimodal adaptive motion control; the intelligent gait planning algorithm employs a composite gait strategy: During meandering motion, based on the Hirose serpentine curve curvature equation: s is the arc length parameter, used to determine the curve position; L is the length of the robot snake, which is involved in the curvature quantization calculation; Dynamically adjust the number of waveforms With the initial swing angle To achieve a minimum bend radius of 0.8 times the pipe diameter, the servo angle function is: For the first Individual servo angle values; For time; Number the joints; During the peristaltic movement, the three wheelless joints arch upwards through a sequence of sinusoidal functions: In the gait switching mechanism, inertial navigation monitors the pipe curvature and diameter in real time, automatically switches the movement mode, and automatically adjusts the gait amplitude according to the pipe material; In friction adaptive control, the roughness of the inner wall is analyzed by visual image texture analysis, and the pressure of the driven wheel and the joint swing amplitude are dynamically adjusted. III. Design a multimodal detection system that integrates vision, thermal imaging, and inertial navigation; the multimodal detection system includes a vision module, a thermal imaging module, and an inertial navigation module; The vision module uses a high-definition fisheye camera and is equipped with the YOLOv5 algorithm to identify cracks in real time. The thermal imaging module uses an infrared sensor with an accuracy of ±1.5℃ to detect temperature anomalies, accurately locate leak points, and solve the problem of missing hidden faults. The inertial navigation module uses a nine-axis IMU to monitor the snake's position and tilt angle in real time, providing reference data for gait planning and defect localization. The multimodal detection system uses a joint detection rule, which employs both visual texture feature extraction and thermal imaging temperature gradient analysis for dual determination. This is combined with an extended Kalman filter to fuse multi-source data, achieving centimeter-level localization. The state equation and observation equation for the extended Kalman filter are as follows: This represents the state vector of the system at time k+1; This is the state transition function; This is the process noise vector; Let be the observation vector at time k; For observation functions; For the observed noise vector; IV. Based on biomimetic robotic snakes, intelligent gait planning algorithms, and multimodal detection systems, the complex working conditions of the inner wall of small-diameter pipelines are detected, including the following stages: 4.

1. Water Entry and Access Phase Connection and debugging: The robotic snake is inserted into the pipe through the inspection port. The tail ballast block is adjusted to neutral buoyancy with a contact force of 5N. The snake head sensor group performs self-tests, camera white balance calibration, and thermal imager temperature calibration. Motion control: A meandering gait with an amplitude of 35° and a frequency of 0.2Hz is adopted to pass through the gate connection. The joint angle function is: Passing through a 150mm narrow section without collision within 2 minutes, with IMU real-time monitoring of attitude deviation <3°, verifying the passability of the orthogonal joint ±85° rotation range; 4.

2. Complex Environment Testing Phase (1) Straight pipe section with moss Gait switching: Upon detection of a stable tube diameter of 200 mm and an inner wall roughness Ra > 5 μm, the gait switches to a peristaltic pattern. The three wheelless joints arch with an amplitude of 40° and a frequency of 0.25 Hz. Joint angle function: The driven wheel adheres to the inner wall with a pressure of 5N and a propulsion speed of 3.5cm / s; (2) 60° bend section with a curvature radius of 240mm Curvature adaptive control: Inertial navigation triggers a meandering turning algorithm, calling the curvature equation. ,in , The joint swing amplitude is dynamically adjusted to 75°. Initial defect screening: The thermal imager detected an abnormal temperature gradient on the inside of the elbow, which was marked as a suspected leakage area, triggering the detailed inspection process; 4.

3. Detailed Inspection Stage for Suspected Defects The snake-head 360° rotating scan: The snake-head servo motor rotates at 5° / second, while a high-definition camera performs macro photography on the suspected area. Combined with edge detection algorithms, the crack is confirmed. Simultaneously, a thermal imager scans and confirms that the temperature in the crack area is 2.5°C lower than the surrounding area, consistent with leakage characteristics. 4.

4. Data Processing and Evacuation Phase Data processing: Detection data is transmitted to the monitoring host via an underwater acoustic communication module. The AI ​​system automatically generates a defect heat map, marking cracks, corrosion areas, and leakage points. Evacuation control: The tail compartment gains 200g of weight and switches to a reverse crawling gait, returning to the maintenance port within 5 minutes, with no scale or blockage in the joints.

Images