Tube wall detection four-footed bionic robot based on magnetic eddy current vision multi-mode fusion

Abstract

The invention relates to the technical field of pipeline nondestructive testing, and discloses a pipe wall detection four-legged bionic robot based on magnetic eddy current vision multi-mode fusion, comprising: a trunk on which an illumination video recording device, a 3D laser radar, an infrared camera, an inertial measurement unit and a control and calculation unit are integrated; the invention relates to a magnetic eddy current detection device, which comprises a movable arm main body, a handle, a mechanical arm and a sensor integrated block, wherein a magnetic eddy current detection probe and a high-resolution near-focus industrial camera are integrated on the sensor integrated block; and the control and calculation unit is electrically connected with the illumination video recording device, the 3D laser radar, the movable arm main body, the infrared camera, the mechanical arm, the inertial measurement unit and each component on the sensor integrated block, and is used for controlling the movement and detection operation of the robot. According to the invention, a four-bar parallel four-foot structure is combined with a passive flexible suspension mechanism, and a multi-modal fusion process of visual fast scanning and magnetic eddy current accurate measurement is adopted, so that efficient and accurate autonomous detection and quantitative evaluation of pipeline defects are realized.

Description

Technical Field

[0001] This invention relates to the field of pipeline non-destructive testing technology, specifically to a quadrupedal biomimetic robot for pipe wall inspection based on magnetic eddy current vision multimodal fusion. Background Technology

[0002] As a critical infrastructure for industrial production and energy transmission, the structural integrity and operational safety of pipelines are of paramount importance. Regular pipeline inspections are necessary to prevent leaks or failures caused by defects such as corrosion and cracks.

[0003] Traditional pipeline inspection methods mainly rely on manual visual inspection or handheld instruments. This approach is not only inefficient and labor-intensive, but also susceptible to subjective factors that can lead to missed detections or misjudgments. More importantly, in-service pipelines are often located in complex and dangerous environments such as high altitudes, enclosed spaces, or containing hazardous media, making manual inspection a direct threat to the safety of workers.

[0004] To overcome the shortcomings of manual inspection, the industry has attempted to apply robotics for automated inspection. However, existing pipeline inspection robots still face technical challenges in practical applications. On the one hand, when the robot moves over uneven surfaces such as curved pipe surfaces, its vibrations are transmitted to the onboard sensors, causing fluctuations in the lift-off value between the sensor and the surface being inspected. This introduces significant signal noise into detection methods sensitive to lift-off values, such as magnetic eddy current detection, directly affecting the accuracy and reliability of defect detection. On the other hand, many existing automation solutions rely on a single detection mode, resulting in limited dimensions of defect information and making it difficult to conduct a comprehensive quantitative assessment of defects. Furthermore, existing technologies lack mature solutions for efficiently combining large-scale rapid inspection with precise local analysis, making it difficult to achieve a balance between efficiency and accuracy in the inspection process.

[0005] Therefore, this invention proposes a quadrupedal bionic robot for pipe wall inspection based on magnetic eddy current vision multimodal fusion to address the shortcomings of existing technologies. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a quadrupedal bionic robot for pipe wall inspection based on magnetic eddy current vision multimodal fusion, which solves the problems of poor motion stability, interference of body vibration with detection signals, and difficulty in balancing inspection efficiency and accuracy in pipeline inspection robots.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a quadrupedal bionic robot for pipe wall inspection based on magnetic eddy current vision multimodal fusion, comprising: The torso integrates a lighting and recording device, a 3D lidar, an infrared camera, an inertial measurement unit, and a control and computing unit. The main body of the movable arm is respectively set at the four corners of the torso, and is used to drive the torso to walk on the surface of the pipe; Handles are used to facilitate the robot's picking up and placing. A robotic arm includes a rotating rod and a movable rod. One end of the rotating rod is connected to the torso via a rotating shaft. The other end of the rotating rod is also equipped with a rotating shaft. Rollers are installed at both ends of the rotating shaft. A sensor integration block is installed on the lower side of the rotating shaft. The sensor integration block integrates a magnetic eddy current detection probe and a near-focus industrial camera. A slide rail is installed at the lower part of the torso. One end of the movable rod is connected to the movable end of the slide rail via a rotating shaft. A shock-absorbing arm is installed at the other end of the movable rod. The lower end of the shock-absorbing arm is connected to the rotating rod via a rotating shaft. The control and computing unit is electrically connected to the lighting and recording device, 3D LiDAR, movable arm body, infrared camera, robotic arm, inertial measurement unit, and various components on the sensor integration block, and is used to control the robot's movement and detection operations.

[0008] Preferably, the main body of the movable arm includes a servo motor one, the output ends of servo motor one are connected to the output ends of servo motor two located at four locations on the torso, the output end of servo motor one is provided with a joint arm two and one end of the upper arm, the other end of the upper arm is provided with a lower leg, the bottom of the lower leg is provided with an electrically controlled permanent magnet, the outer side of the upper arm is provided with a joint arm three, the outer side of servo motor one is provided with one end of joint arm one, the other end of joint arm one is provided with a connecting rod one between joint arm two, and the joint arm two is provided with a connecting rod two between joint arm two and joint arm three.

[0009] Preferably, an electrically controlled permanent magnet is provided at the bottom of the lower leg; the electrically controlled permanent magnet is electrically connected to the control and computing unit, and the control and computing unit provides the robot with a controllable adsorption force on the surface of the metal pipe by controlling the magnetization and demagnetization of the electrically controlled permanent magnet; for non-metallic pipes, adsorption can be achieved by increasing the friction at the foot end, that is, the adsorption surface of the electrically controlled permanent magnet is surrounded by a high friction coefficient anti-slip material.

[0010] Preferably, the linkage mechanism of the robotic arm is a passive compliant suspension mechanism; the passive compliant suspension mechanism utilizes the geometric constraints of the linkage and gravity to provide a stable support force for the sensor integrated block without active force control when the roller contacts the pipe surface, so as to maintain a constant lift-off value between the magnetic eddy current detection probe and the pipe surface when the robot body vibrates.

[0011] Preferably, when the rollers of the passive compliant suspension mechanism are in contact with the pipe surface, the control and computing unit does not perform active force control on the robotic arm, and the supporting force obtained by the sensor integrated block... The supporting force is determined by the inherent physical characteristics of the passive compliant suspension mechanism. The mechanical model is as follows: ; In the formula, This is the equivalent spring constant of the shock absorber arm in a passive compliant suspension system. This is the equivalent damping coefficient. This refers to the compression displacement of the passive compliant suspension mechanism. The compression velocity is the value of the passive compliant suspension mechanism; the mechanical model enables the passive compliant suspension mechanism to effectively absorb high-frequency vibrations from the torso.

[0012] Preferably, the magnetic eddy current detection probe includes an excitation coil and a pickup unit; the excitation coil is used to generate a main magnetic field inside the pipe wall, and the main magnetic field generates a leakage magnetic field when it encounters a volumetric defect, and induces an eddy current field on the pipe surface; the pickup unit is used to simultaneously detect the composite magnetic field signal composed of the leakage magnetic field and the eddy current field modulated by the surface defect.

[0013] Preferably, the pickup unit adopts a differential structure, including a pickup unit one and a pickup unit two symmetrically arranged; the control and calculation unit receives the differential output voltage. for: ; In the formula, To pick up the induced voltage of unit one, To pick up the induced voltage of unit two; The differential structure is used to suppress common-mode noise caused by lift-off value fluctuations and temperature changes, thereby improving the signal-to-noise ratio of the defect signal.

[0014] Preferably, the control and computing unit is further configured as follows: By integrating the environmental 3D point cloud data acquired by the 3D LiDAR with the robot posture data acquired by the inertial measurement unit, and by running an instantaneous localization and mapping algorithm, a pipeline environment map is generated in real time and the robot is located to achieve autonomous navigation.

[0015] Preferably, the control and computing unit is further configured as follows: The differential output voltage After preprocessing, the data is input into a pre-trained neural network model. The output of the neural network model is a quantitative assessment result of pipeline defects, which includes defect category and defect depth.

[0016] Preferably, the control and computing unit is further configured as follows: Execute the paused scan detection workflow, the workflow including: The robot is controlled to walk to the target detection area while the robotic arm is retracted. Stop walking; Control the robotic arm to descend until the rollers stably contact the pipe surface; Perform a local scan detection.

[0017] This invention provides a quadrupedal bionic robot for pipe wall inspection based on magnetic eddy current vision multimodal fusion, which has the following beneficial effects: 1. This invention improves the robot's motion stability and load-bearing capacity in unstructured environments such as curved pipe surfaces by employing a four-bar parallel quadruped inspection robot mechanical structure and combining it with a multi-gait planning algorithm that considers the crawling requirements of pipe inspection. This combination of structural design and control algorithm enables the robot platform to not only stably carry multiple inspection modules but also achieve flexible obstacle avoidance and environmental adaptation, providing a reliable mobile platform foundation for subsequent precise inspection operations.

[0018] 2. This invention deploys the detection sensor using a passive compliant suspension mechanism. This mechanism utilizes the inherent physical properties of the machine to effectively isolate vibrations transmitted from the robot's torso during movement, ensuring a constant lift-off value between the magnetic eddy current detection probe and the measured surface. Combined with the probe's differential signal pickup structure, this suppresses noise interference during the detection process, thereby enabling the stable acquisition of high-quality magnetic eddy current coupled detection signals and solving the key technical problem of signal susceptibility to interference in mobile detection platforms.

[0019] 3. This invention establishes a multimodal fusion-based autonomous detection and evaluation process. This process utilizes machine vision and an improved neural network algorithm for rapid autonomous exploration and preliminary defect identification of pipelines, followed by high-precision quantitative prediction using magnetic eddy current composite signals. This strategy, from rapid macroscopic inspection to precise microscopic quantitative analysis, automates the entire pipeline defect detection process, improving not only detection efficiency but also the accuracy and reliability of defect assessment through the fusion and intelligent interpretation of multi-source information. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall structure of the present invention; Figure 2 The diagram shows the overall structure of the invention from a bottom view. Figure 1 ; Figure 3 This is a left-side view of the overall structure of the present invention; Figure 4 This is a front view schematic diagram of the overall structure of the present invention; Figure 5 This is a top view of the overall structure of the present invention; Figure 6 The diagram shows the overall structure of the invention from a bottom view. Figure 2 ; Figure 7 This is a flowchart of the method of the present invention; Figure 8 This is a block diagram of the functional modules and data flow of the present invention.

[0021] The components include: 1. Torso; 2. Lighting and recording device; 3. 3D LiDAR; 4. Main body of the movable arm; 401. Servo motor one; 402. Upper arm; 403. Lower leg; 404. Electro-controlled permanent magnet; 405. Connecting rod one; 406. Connecting rod two; 407. Articulated arm one; 408. Articulated arm two; 409. Articulated arm three; 5. Handle; 6. Mechanical arm; 601. Rotating rod; 602. Movable rod; 603. Shock-absorbing arm; 604. Roller; 605. Sensor integrated block; 7. Slide rail; 8. Infrared camera; 9. Servo motor two. Detailed Implementation

[0022] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0023] Reference Figures 1 to 6 This invention provides a quadrupedal biomimetic robot for pipe wall inspection based on magnetic eddy current vision multimodal fusion. The robot's overall structure includes a torso 1, four sets of movable arm bodies 4 located at the four corners of the torso 1, and a set of mechanical arms 6 located at the lower part of the torso 1.

[0024] The torso 1 serves as the core platform for the robot, integrating a 3D LiDAR 3 for three-dimensional environmental perception, an infrared camera 8 for auxiliary observation, an illumination and recording device 2, and an inertial measurement unit for attitude measurement. A handle 5 is located on the top of the torso 1 for transporting and placing the entire robot. All data processing and control commands for the robot are issued by the control and computing unit located inside the torso 1.

[0025] Four sets of movable arms 4 constitute the robot's mobile platform, used to drive the robot to walk on surfaces such as pipes. Each set of movable arms 4 is driven by a servo motor and has a four-bar parallel configuration to improve the robot's stability when moving on irregular curved surfaces. The ends of the lower legs 403 are equipped with electrically controlled permanent magnets 404, which provide controllable adsorption force for the robot to walk on ferromagnetic pipes by magnetizing and demagnetizing.

[0026] The robotic arm 6 is a passive compliant suspension mechanism with a sensor integration block 605 connected to its end. The sensor integration block 605 integrates a magnetic eddy current detection probe and a high-resolution near-focus industrial camera. The function of this mechanism is to stably attach the sensor integration block 605 to the pipe surface when the robot is stationary for inspection, and to isolate the robot body from vibration through its own physical properties, ensuring the quality of the inspection data.

[0027] Reference Figure 7 This invention also provides a pipe wall detection method based on the above-described robot. This method is executed by a control and computing unit and may include the following steps: S100: Environmental perception and motion control. After the robot starts, it integrates data from the 3D LiDAR 3 and the inertial measurement unit, runs a real-time localization and mapping algorithm to generate an environmental map and perform self-localization; at the same time, it runs a multi-gait planning algorithm to control the four sets of movable arms 4 to drive the robot to walk, turn or avoid obstacles autonomously according to the planned path.

[0028] S200: Visual Inspection and Initial Defect Localization. During robot movement, a high-resolution near-focus industrial camera located on sensor integration block 605 continuously acquires images of the pipe surface. The control and computing unit processes the acquired images in real time, quickly identifying macroscopic defects such as cracks and corrosion on the pipe surface by running a lightweight target detection neural network model, and recording their location coordinates on the map.

[0029] S300: Stop-and-Scan Precision Measurement and Quantitative Assessment. When visual inspection detects a suspected defect or reaches a preset detection point, the robot executes a stop-and-scan workflow. The robot stops moving, and the robotic arm 6 descends, causing the sensor integrated block 605 to stably contact the pipe surface. Subsequently, the magnetic eddy current detection probe scans the target area, acquiring composite magnetic field signals. Finally, the acquired signals are input into a quantitative analysis neural network model, outputting a quantitative assessment result of the defect type and depth.

[0030] See attached document Figure 8 To implement the above method, the control and computing unit may logically include multiple functional modules, including: The environmental perception and positioning module a is used to process sensor data from the 3D LiDAR 3 and the inertial measurement unit; Motion control module b is used to generate and execute multi-gait walking commands for the robot; Visual detection module C is used to run the target detection neural network model and process camera images; The magnetic eddy current analysis module d is used to process probe signals and run a quantitative analysis neural network model. The task scheduling and process control module e is used to coordinate the operation of all other modules and manage the overall workflow from S100 to S300.

[0031] Reference Figures 1 to 6 This section will elaborate on the core mechanical components that make up this quadrupedal bionic robot, including its mobile platform, sensor deployment mechanism, and sensor integration method.

[0032] Reference Figure 1 , Figure 3 , Figure 4 , Figure 5 The robot's mobile platform is centered on its torso 1. The torso 1 houses a control and computing unit and an inertial measurement unit, while externally it integrates a 3D LiDAR 3 for acquiring three-dimensional environmental information and an infrared camera 8 for assisting observation.

[0033] Reference Figure 1 and Figure 2 Four sets of movable arms 4 are symmetrically arranged at the four corners of the torso 1, together forming the robot's walking mechanism. Each set of movable arms 4 consists of servo motor 1 401, servo motor 2 9, upper arm 402, lower leg 403, articulated arm 1 407, articulated arm 2 408, articulated arm 3 409, connecting rod 1 405, and connecting rod 2 406. These components are connected by a pivot, forming a four-bar parallel structure. Compared with the series configuration, this parallel configuration has higher structural rigidity and load-bearing capacity, and its kinematic characteristics enable the robot to maintain the stability of the torso 1 when walking on irregular curved surfaces such as pipes.

[0034] Reference Figure 2 , Figure 3 , Figure 6 The robot's torso 1 has a robotic arm 6 mounted on its lower part. This robotic arm 6 is a passive compliant suspension mechanism. The mechanism includes a rotating rod 601, one end of which is connected to the torso 1 via a pivot. A slide rail 7 is also located on the lower part of the torso 1. One end of a movable rod 602 is connected to the movable end of the slide rail 7 via a pivot. The other end of the movable rod 602 is connected to a shock-absorbing arm 603, and the lower end of the shock-absorbing arm 603 is connected to the middle of the rotating rod 601 via a pivot, together forming a linkage mechanism.

[0035] During the detection task, the mechanism unfolds under gravity until its end roller 604 contacts the pipe surface. At this time, the control and computing unit does not perform any active force control on the mechanism. Due to its own weight and geometric constraints, the mechanism generates a stable supporting force on the end sensor integrated block 605. The physical characteristics of the mechanism cause its linkage system to generate corresponding compressive displacement and velocity when subjected to vibration from the torso 1, thereby generating a dynamic supporting force. The mechanical model of this supporting force can be approximated as a spring-damped system, characterized by the following equation: ; In the formula, This is the equivalent spring constant of the shock absorber arm in a passive compliant suspension system. This is the equivalent damping coefficient, which is determined by factors such as friction at the connections of the various rotating shafts in the mechanism. This refers to the compression displacement of the passive compliant suspension mechanism. The compression speed of the passive compliant suspension mechanism; this mechanical property enables the mechanism to passively absorb high-frequency vibrations generated by the torso 1 during walking, thereby maintaining a constant lift-off value between the sensor integrated block 605 and the pipe surface.

[0036] See attached document Figure 2 and Figure 6 The sensor integrated block 605, fixed to the end of the rotating rod 601, is the core sensing unit for performing multimodal detection. The housing of this integrated block houses a magnetic eddy current detection probe and a high-resolution near-focus industrial camera. The optical axis of the high-resolution near-focus industrial camera is approximately perpendicular to the pipe surface, used to acquire high-definition images of the pipe surface.

[0037] To ensure clear imaging inside poorly lit pipes, the robot is equipped with a lighting system. In one embodiment, an infrared camera 8 has an external LED light mounted on its torso 1, with its illumination directed towards the detection area in front of the sensor integration block 605. In another embodiment, a ring-shaped LED light source array can be integrated around the lens of a high-resolution near-focus industrial camera on the sensor integration block 605 to provide uniform, shadowless illumination for visual inspection.

[0038] The roller 604 is mounted together with the sensor integration block 605 on the shaft at the end of the rotating rod 601. When the robotic arm 6 extends, the roller 604 first contacts the pipe surface. Since the radius of the roller 604 is fixed, its function is to accurately define the distance between each sensor on the sensor integration block 605 (especially the magnetic eddy current detection probe) and the pipe surface, i.e., the lift-off value. At the same time, during partial scanning detection, the roller 604 enables the sensor integration block 605 to roll smoothly along the pipe surface, reducing friction and avoiding damage to the sensors or the pipe surface.

[0039] Reference Figure 7 and Figure 8 This section will elaborate in detail on the physical principles and fusion methods of the two core detection modes used in the embodiments of the present invention.

[0040] The operation of the eddy current detection probe is based on the combination of two physical effects: electromagnetic induction and magnetic field distortion. The probe internally includes an excitation coil and a pickup unit. During operation, the control and computing unit drives the excitation coil to receive an alternating current of a specific frequency, generating an alternating main magnetic field in the ferromagnetic pipe wall material. According to Faraday's law of electromagnetic induction, this alternating main magnetic field induces a closed-loop eddy current field on the pipe surface.

[0041] When there are volumetric defects such as cracks or corrosion on or near the surface of a pipe, two physical phenomena will occur.

[0042] Firstly, because the magnetic resistance at the defect is much greater than that of intact metal, the magnetic field lines of the main magnetic field will be distorted, and some magnetic field lines will penetrate out of the pipe surface, forming a leakage magnetic field above the defect.

[0043] Secondly, the presence of defects can obstruct the normal flow path of induced eddies, causing them to flow around or be interrupted, thus altering the distribution of the eddy field around the defect.

[0044] The pickup unit is a high-sensitivity magnetic sensor array designed to simultaneously detect the composite magnetic field signal generated by the two phenomena mentioned above. That is, the pickup unit simultaneously senses the leakage magnetic field signal caused by the defect, as well as the secondary magnetic field signal generated by the change in the eddy current field.

[0045] To improve the signal-to-noise ratio, the pickup unit in this embodiment adopts a differential structure. This structure includes two symmetrically arranged pickup units, Unit 1 and Unit 2. During detection, these two pickup units are placed adjacent to each other in the area being measured. When there are no defects, the background magnetic fields sensed by the two pickup units are essentially the same; when one pickup unit sweeps over a defect, its induced voltage changes significantly. The control and calculation unit receives the induced voltages from the two pickup units and performs differential calculations. The obtained differential output voltage is characterized by the following formula: ; In the formula, The differential output voltage is received by the control and calculation unit and used for subsequent analysis. The induced voltage of the pickup unit at a certain moment; This represents the induced voltage of pickup unit two at the same time. Through this differential operation, the common-mode noise signals acting simultaneously on the two pickup units, caused by factors such as lift-off value fluctuations, temperature drift, or external electromagnetic interference, will cancel each other out, thereby effectively improving the detection sensitivity of defect signals.

[0046] The visual detection technology solution in this embodiment achieves rapid processing of images captured by a high-resolution close-focus industrial camera by deploying a specifically optimized target detection neural network model in the control and computing unit. This model can be built based on a mature target detection architecture (such as YOLOv5).

[0047] To accommodate the limited computing resources of the robot's embedded computing platform and meet the real-time requirements of inspection tasks, a Group Spatial Convolution (GSConv) module is introduced into the network structure of this model. The GSConv module is a structural unit that reduces the computational complexity of convolutional neural networks. Its core principle is to decompose the standard convolution operation. For an input feature map, the GSConv module first divides its channels into two parts. One part is processed by standard convolution to retain rich feature information; the other part is processed by depthwise separable convolution, which has lower computational cost. Subsequently, the results of the two processing parts are concatenated, and a channel-shuffle operation is used to promote the interaction and fusion of feature information from the two parts.

[0048] By replacing some standard convolutional layers in the original network with the GSConv module, the number of floating-point operations (FLOPs) and parameters can be significantly reduced while maintaining high detection accuracy. This design reduces the model's computational resource consumption and improves its inference speed in the robot control and computing unit, thereby achieving efficient and real-time recognition of pipeline surface defect images.

[0049] See attached document Figure 7 and attached Figure 8 This section will provide a detailed description of the internal logic functions of the control and computing unit and the fully autonomous detection workflow executed by it.

[0050] The robot's autonomous movement is performed by the control and computing unit. First, the environmental perception and localization module 'a' fuses multi-source data from external sensors. Specifically, this module receives 3D point cloud data of the environment from the 3D LiDAR 3, as well as real-time robot attitude data (including roll, pitch, and yaw angles) from the built-in inertial measurement unit. Based on the fused data, this module runs a Simultaneous Localization and Mapping (SLAM) algorithm to generate a digital map of the pipeline environment in which the robot is located in real time, and continuously tracks the robot's precise position and attitude on this map.

[0051] After obtaining positioning and map information, motion control module b plans an optimal path from the current location to the target location on the map based on preset tasks or instructions from other modules. Subsequently, this module executes the vehicle motion control (VMC) algorithm, decomposing the planned path into a series of specific motion commands. This algorithm further calculates the precise motion parameters of each servo motor driving the four sets of movable arm bodies 4 through multi-gait planning, in order to control the four-bar parallel leg structure to achieve complex movements, such as stable walking on straight pipe sections, posture adjustment at curved pipes, and specific gaits for crossing obstacles such as flanges.

[0052] The detection method in this embodiment adopts a two-level workflow, which is coordinated by the task scheduling and process control module e, in order to balance detection efficiency and accuracy.

[0053] Phase 1: Perform rapid visual inspection. As the robot autonomously moves along the pipeline, a high-resolution near-focus industrial camera mounted on sensor IC 605 continuously acquires images of the pipeline surface. The acquired image stream is sent to the vision inspection module c in real time. This module c runs a lightweight target detection neural network model integrating the GSConv module to analyze the images and identify macroscopic defects such as cracks and corrosion spots on the pipeline surface. Once a suspected defect is identified, the module records the image features of the defect and its location coordinates in the SLAM map, completing the initial defect localization.

[0054] The second stage involves a stop-and-scan fine-tuning inspection. Once the vision inspection module c locates a suspected defect, the task scheduling and process control module e triggers the second-stage workflow. The motion control module b controls the robot to precisely move to the recorded defect coordinates and stop. Subsequently, the control and computing unit controls the robotic arm 6 to unfold. This arm, a passive compliant suspension mechanism, descends under gravity until the roller 604 stably contacts the pipe surface, thus deploying the sensor integration block 605. After deployment, the magnetic eddy current analysis module d is activated, controlling the magnetic eddy current detection probe to perform a high-precision scan of the small localized area where the defect is located and to collect the composite magnetic field signal of that area.

[0055] This step is the final step in data processing and analysis. The differential output voltage signal acquired by the magnetic eddy current detection probe during the second stage of scanning is transmitted to the magnetic eddy current analysis module d. This module first preprocesses the signal, including digital filtering, noise reduction, and normalization, to extract signal segments that can effectively characterize defect features.

[0056] Subsequently, the preprocessed signal data is input into a pre-trained quantitative analysis neural network model. This model, trained on a large number of defect sample signals of known types and sizes, has internal weights capable of characterizing the complex nonlinear mapping relationship between signal features and defect physical parameters. Therefore, upon receiving a new signal input, the model can directly output a quantitative assessment of the defect, including the defect category (e.g., crack, pitting) and key geometric parameters, such as defect depth.

[0057] This section will summarize how the above technical modules work together and explain the comprehensive technical effects achieved when they are integrated into a complete system.

[0058] Reference Figures 1 to 8 This invention integrates specific mechanical structures with multimodal sensing and intelligent algorithms to form a closed-loop autonomous detection system. The entire system is coordinated by a control and computing unit, whose internal task scheduling and process control module e is responsible for managing and executing the complete detection tasks.

[0059] The system's operation begins with autonomous movement. The environmental perception and localization module a integrates data from the 3D LiDAR 3 and the inertial measurement unit, providing the motion control module b with high-precision real-time localization and an environmental map. Based on this information, the motion control module b performs path planning and multi-gait control, driving the movable arm 4, which consists of a four-bar parallel structure, enabling the robot to move stably and autonomously in the complex pipe environment. The combination of structure and algorithm in this stage provides reliable mobile execution capabilities for all subsequent detection tasks.

[0060] During the movement, the system executes a two-stage workflow of visual inspection and magnetic eddy current precision measurement. In the first stage, the visual inspection module c performs real-time analysis on images acquired by a high-resolution near-focus industrial camera. Due to the adoption of the GSConv structure in the model, this visual analysis process has low computational complexity while ensuring recognition accuracy, meeting the efficiency requirements of real-time inspection on the mobile platform. When the visual inspection module c identifies a suspected defect location, the task scheduling and process control module e initiates the second stage. The robot moves to the target point and stops. The robotic arm 6, i.e., the passive compliant suspension mechanism, uses its own physical characteristics to stably deploy the sensor integrated block 605 onto the pipe surface without the need for active power control, effectively isolating the influence of residual vibration of the robot body on the detection.

[0061] Once deployed, the magnetic eddy current analysis module d begins operation, acquiring composite magnetic field signals from the defect area. These signals are then input into a quantitative analysis neural network model to accurately assess the defect type and depth. This completes a full detection process, from large-scale autonomous surveying and rapid visual localization to precise local scanning and final quantitative analysis.

[0062] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A pipe wall detection quadruped robot based on magnetic eddy current vision multimodal fusion, characterized in that, It comprises: a trunk (1) integrated with an illumination video device (2), a 3D laser radar (3), an infrared camera (8), an inertial measurement unit, and a control and calculation unit; a movable arm body (4) arranged at each corner of the trunk (1) for driving the trunk (1) to walk on the surface of a pipeline; a handle (5) for facilitating the taking and placing of the robot; a mechanical arm (6) comprising a rotating rod (601) and a movable rod (602), one end of the rotating rod (601) being connected to the trunk (1) through a rotating shaft, the other end of the rotating rod (601) being provided with a rotating shaft, both ends of the rotating shaft being provided with a roller (604), the lower side of the rotating shaft being provided with a sensor integrated block (605), the sensor integrated block (605) being integrated with a magnetic eddy current detection probe and a near-focus industrial camera, the lower part of the trunk (1) being provided with a sliding rail (7), one end of the movable rod (602) being connected to the movable end of the sliding rail (7) through a rotating shaft, the other end of the movable rod (602) being provided with a shock absorbing arm (603), the lower end of the shock absorbing arm (603) being connected to the rotating rod (601) through a rotating shaft; the control and calculation unit is electrically connected with the illumination video device (2), the 3D laser radar (3), the movable arm body (4), the infrared camera (8), the mechanical arm (6), the inertial measurement unit, and each component on the sensor integrated block (605) for controlling the movement and detection work of the robot.

2. The four-legged bionic robot for pipeline wall detection based on magnetic eddy current vision multi-modal fusion according to claim 1, wherein the movable arm body (4) comprises a servo motor one (401), the output end of the servo motor one (401) being connected to the output end of a servo motor two (9) arranged at four places of the trunk (1), the output end of the servo motor one (401) being provided with a joint arm two (408) and one end of a large arm (402), the other end of the large arm (402) being provided with a small leg (403), the bottom of the small leg (403) being provided with an electrically controlled permanent magnet (404), the outer side of the large arm (402) being provided with a joint arm three (409), one end of a joint arm one (407) being provided on the outer side of the servo motor one (401), a connecting rod one (405) being arranged between the other end of the joint arm one (407) and the joint arm two (408), and a connecting rod two (406) being arranged between the joint arm two (408) and the joint arm three (409).

3. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 2, characterized in that, The bottom of the small leg (403) is provided with an electrically controlled permanent magnet (404); the electrically controlled permanent magnet (404) is electrically connected with the control and calculation unit, the control and calculation unit controls the magnetization and demagnetization of the electrically controlled permanent magnet (404) to provide controllable adsorption force for the robot on the surface of a metal pipeline; for a non-metal pipeline, adsorption is achieved by increasing the friction force of the foot end, and the adsorption surface of the electrically controlled permanent magnet (404) is surrounded by a non-slip material.

4. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 1, characterized in that, The linkage mechanism of the mechanical arm (6) is a passive compliant suspension mechanism; the passive compliant suspension mechanism utilizes linkage geometry constraints and gravity to provide a stable support force for the sensor integrated block (605) when the roller (604) contacts the pipeline surface, so as to keep the lift-off value between the magnetic eddy current detection probe and the pipeline surface constant when the robot body vibrates.

5. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 4, characterized in that, The passive compliant suspension mechanism, when the roller (604) is in contact with the pipeline surface, the control and calculation unit does not actively control the mechanical arm (6), and the support force obtained by the sensor integrated block (605) The support force The mechanical model of the passive compliant suspension mechanism is determined by the physical characteristics of the passive compliant suspension mechanism. ； wherein, is the equivalent spring coefficient of the shock absorbing arm (603) of the passive compliant suspension mechanism, is the equivalent damping coefficient, is the compression displacement of the passive compliant suspension mechanism, is the compression velocity of the passive compliant suspension mechanism; the mechanical model enables the passive compliant suspension mechanism to effectively absorb high frequency vibrations from the torso (1).

6. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 1, characterized in that, The magnetic eddy current detection probe includes an excitation coil and a pickup unit; the excitation coil is used to generate a main magnetic field in the pipeline wall, the main magnetic field generates a leakage magnetic field when encountering a volumetric defect, and induces an eddy current field on the pipeline surface; the pickup unit is used to synchronously detect a composite magnetic field signal composed of the leakage magnetic field and the eddy current field modulated by the surface defect.

7. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 6, characterized in that, The pickup unit adopts a differential structure, comprising pickup unit one and pickup unit two symmetrically arranged; the control and calculation unit receives differential output voltage is: ； wherein is the induced voltage of the pickup unit one, is the induced voltage of the pickup unit two; The differential structure is used to suppress common-mode noise generated by lift-off value fluctuation and temperature change, and improve the signal-to-noise ratio of the defect signal.

8. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 1, characterized in that, The control and calculation unit is further configured to: fuse the environment three-dimensional point cloud data acquired by the 3D laser radar (3) and the robot pose data acquired by the inertial measurement unit, generate a pipeline environment map in real time and position the robot by running a real-time positioning and map construction algorithm, so as to realize autonomous navigation.

9. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 7, characterized in that, The control and calculation unit is further configured to: The differential output voltage After pre-processing, the input is input to a pre-trained neural network model, and an output of the neural network model is a quantitative evaluation result of the pipeline defect, and the quantitative evaluation result includes a defect category and a defect depth.

10. The pipe wall inspection quadruped robot based on magnetic eddy current vision multi-modal fusion according to claim 4, characterized in that, The control and calculation unit is further configured to: execute a stop scanning detection workflow, the workflow including: controlling the robot to walk to a target detection area in a state where the mechanical arm (6) is retracted; stop walking; control the mechanical arm (6) to descend until the roller (604) stably contacts the pipeline surface; execute local scanning detection.

Images