Pipeline detection method and system based on intelligent pipeline robot with body

Abstract

The invention discloses a pipeline detection method and system based on an intelligent pipeline robot with a body, and relates to the technical field of pipeline detection. The method comprises the following steps: acquiring pose, environment and pipe wall sensing data acquired by a multi-mode sensor; a pipeline three-dimensional digital map is constructed, and real-time self-positioning is achieved; identifying a pipe wall defect based on the pipe wall sensing data, generating a defect data packet, and binding the defect data packet with a map space position for storage; when an adjacent robot is detected, constructing a temporary communication link; exchanging detection information containing a map abstract and task state information with an adjacent robot through the link; updating the three-dimensional digital map by using the map abstract of the adjacent robot; performing task negotiation by combining the task state information and the updated map; and adjusting the target operation area and the motion path according to the negotiation result. According to the invention, full-automatic high-precision detection and intelligent operation and maintenance decision-making in a complex pipeline environment can be realized.

Description

Technical Field

[0001] This invention relates to the field of pipeline inspection technology, and in particular to a pipeline inspection method and system based on an embodied intelligent pipeline robot. Background Technology

[0002] Currently, pipeline inspection operations face significant technical bottlenecks when dealing with complex and unknown pipeline environments. Traditional inspection methods heavily rely on manual operation and pre-acquired pipeline drawings, making them unsuitable for unfamiliar pipeline scenarios lacking prior knowledge. Existing pipeline robots, upon entering unknown pipelines, generally lack real-time, high-precision autonomous localization and mapping capabilities, hindering truly unmanned autonomous operation.

[0003] Furthermore, traditional single-robot inspection methods struggle to handle large-scale detection tasks, resulting in low inspection efficiency and failing to meet the demands of efficient pipeline network operation and maintenance. While existing technologies have explored multi-robot collaborative operations, these systems generally fail in unfamiliar pipeline scenarios lacking pre-deployed communication networks. Specifically, upon entering unknown pipelines, multiple robots lack the necessary communication and coordination capabilities. Each robot relies solely on its own sensors for environmental perception and decision-making, creating information silos and frequently leading to repeated detection of the same area, resulting in resource waste. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a pipeline inspection method and system based on an embodied intelligent pipeline robot, which can realize fully automatic high-precision inspection and intelligent operation and maintenance decision-making in complex pipeline environments.

[0005] To achieve the above objectives, the present invention provides a pipeline inspection method based on embodied intelligent pipeline robots, wherein the number of pipeline robots is multiple, and each robot is equipped with a multimodal sensor and a local communication module; the method is executed by each of the pipeline robots and includes: Acquire the pose data, internal environmental data of the pipeline, and pipe wall sensing data collected by the multimodal sensor; Based on the pose data and environmental data, a three-dimensional digital map of the pipeline is constructed and real-time self-localization is achieved; Based on the pipe wall sensing data, pipe wall defects in the pipeline are identified, corresponding defect data packages are generated, and they are bound and stored with the spatial location information in the three-dimensional digital map. When a neighboring robot is detected through the local communication module, a temporary communication link is established; the neighboring robot is another pipeline robot located within its preset communication range. Based on the temporary communication link, they exchange their respective detection information with the neighboring robots; the detection information includes map summary data and task status information. The 3D digital map is updated based on the map summary data of the neighboring robots; Based on the task status information of the neighboring robots and the updated 3D digital map, conduct task negotiation with the neighboring robots; Based on the results of the task negotiation, adjust your target work area and movement path.

[0006] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: The pipeline inspection method based on an embodied intelligent pipeline robot provided by this invention acquires pose, environment, and pipe wall perception data collected by multimodal sensors, and autonomously constructs a three-dimensional digital map and achieves real-time self-localization in unfamiliar pipeline environments. This eliminates the need for the robot to rely on pre-acquired pipeline drawings, significantly improving its environmental adaptability and autonomous operation capabilities in pipeline scenarios lacking prior information.

[0007] When a neighboring robot is detected in the pipeline, a temporary communication link is established to exchange detection information, including map summaries and task status information, effectively breaking down information silos between multiple robots. This allows each robot to access the detection results and status of its companions, creating conditions for collaborative operation. The robot updates its own 3D digital map using map summary data from neighboring robots and negotiates tasks with the updated map, combining this with its task status information. Based on the negotiation results, the robot adjusts its target work area and movement path, enabling multiple robots to work as a coordinated whole. This intelligent allocation of detection tasks at the system level completely avoids the common phenomenon of repeated detection of the same area in traditional multi-robot systems, significantly improving the overall detection efficiency and resource utilization of the vast pipeline network. Attached Figure Description

[0008] Figure 1 This is a schematic flowchart illustrating the pipeline inspection method based on an embodied intelligent pipeline robot according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the detection preparation process shown in an embodiment of the present invention; Figure 3 This is a schematic diagram of the robot self-inspection process shown in an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of the pipeline robot shown in an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware platform architecture shown in an embodiment of the present invention; Figure 6 This is a software function flowchart shown in an embodiment of the present invention; Figure 7 This is a hardware-algorithm collaborative architecture diagram shown in an embodiment of the present invention; Figure 8 This is a schematic diagram of pipeline defect detection as shown in an embodiment of the present invention; Figure 9 This is a schematic diagram illustrating the obstacle avoidance process in an embodiment of the present invention; Figure 10 This is a schematic diagram of the control strategy shown in an embodiment of the present invention; Figure 11 This is a closed-loop diagram of digital twin operation and maintenance as shown in an embodiment of the present invention; Figure 12 This is a flowchart illustrating the post-processing and digital twin synchronization process of pipeline robot inspection data in an embodiment of the present invention. Figure 13 This is a schematic diagram of the modular structure of a pipeline inspection system based on an embodied intelligent pipeline robot, as shown in an embodiment of the present invention. Detailed Implementation

[0009] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0010] Please see Figure 1 , Figure 1 This is a schematic diagram of the pipeline inspection method based on an embodied intelligent pipeline robot.

[0011] This invention provides a pipeline inspection method based on embodied intelligent pipeline robots (hereinafter referred to as "pipeline robots" or "robots"), wherein there are multiple pipeline robots, each equipped with a multimodal sensor and a local communication module; the method is executed by any one of the pipeline robots and includes: S101: Acquire pose data, internal environmental data of the pipeline, and pipe wall sensing data collected by multimodal sensors.

[0012] See Figure 2 During the preparation phase, the robot first checks the basic parameters of the pipeline, such as diameter, material, and bending angle, to determine its compatibility. If it doesn't match, the detection stops and a notification is displayed; if it matches, it removes large debris and accumulated water from the pipeline if necessary. Next, it checks the pipeline environment for high temperature, high humidity, or corrosive gases. If these are present, it activates a protective mode, i.e., a moisture-proof / corrosion-proof mode; otherwise, it enters the default mode and loads the corresponding detection module based on the pipeline material. Metal pipelines are equipped with eddy current sensors, while non-metallic pipelines only use a vision module. The pipeline robot then begins operation.

[0013] See Figure 3During startup, the robot first performs a hardware self-check, including battery level detection and loading of a dynamic energy consumption model. If the battery level falls below the dynamic calculation threshold, an alarm is triggered and startup terminates; otherwise, it continues. Next, it performs a sensor self-check, completing vision / LiDAR / IMU calibration, communication module testing, verifying only the local cache function, and testing the linkage between the motor / steering wheel / guide ball. Any abnormality in any of these steps triggers an alarm and terminates startup; if all steps are normal, it proceeds to the next step. Subsequently, the edge computing platform loads the domain-adapted large model and small-sample learning weight file, performing model initialization verification to match the detection parameters of the current pipeline material. If verification fails, an alarm is triggered indicating a domain-adaptation error; if verification passes, the robot is deployed into the pipeline, activating self-driving mode.

[0014] After the pipeline inspection task begins, the embodied intelligent pipeline robot comprehensively collects environmental data and pipe wall perception data from its multimodal sensors. Environmental data primarily includes 3D point cloud information generated in real-time by LiDAR scanning, with a horizontal field of view of at least 360° and a vertical field of view of at least 50°, and a point cloud density of at least 100 points per square meter, accurately reconstructing the internal geometry and spatial obstacle distribution of the pipeline. Simultaneously, a high-resolution camera acquires images of the pipeline's inner wall at fixed intervals, equipped with an LED lighting module and an adaptive polarization filtering algorithm, ensuring clear images even in low light conditions or under interference from oil stains or water mist, effectively supporting visual SLAM and defect identification. Pose data is provided collaboratively by an inertial measurement unit (IMU) and a high-precision wheel encoder, jointly constituting the robot's motion state and position / attitude information. Pipe wall perception data relies on an ultrasonic sensor to detect pipe wall thickness in real-time at a sampling rate of 500Hz, with an accuracy of ±0.1mm, applicable to pipes of various materials, including metals and non-metals.

[0015] The sensors are adaptable to different pipe materials and operating conditions. Eddy current sensors are used in metal pipes to enhance the detection sensitivity of corrosion defects, while in non-metallic pipes, visual and ultrasonic data are used for comprehensive judgment. The sensor probe surface is covered with a nano-anti-fouling coating with a contact angle of no less than 110°. Combined with adaptive polarization filtering, it can improve image contrast by more than 40% even when oil coverage is less than 30%, effectively suppressing the impact of reflection and dirt adhesion on image quality. For oily or high-humidity environments, a 1550nm wavelength LiDAR is selected, which has superior penetration capability compared to conventional 905nm equipment, keeping the point cloud data loss rate below 8% and ensuring the continuity and integrity of the point cloud map.

[0016] The entire sensing system can be encapsulated with polytetrafluoroethylene (PTFE) material, and the circuit interface meets the IP68 waterproof rating. It can work stably in harsh environments with a humidity of 95%RH and a hydrogen sulfide concentration of no more than 500ppm, and can even withstand immersion in 0.5 meters of water for 24 hours without affecting communication and data acquisition functions.

[0017] See Figure 4 , Figure 4 This is a schematic diagram of the pipeline robot structure; Figure 4 middle, and Represents proactive driving, , and The numbers 1 and 2 represent the transmission drive, respectively, indicating the tension spring and the passive rotation.

[0018] like Figure 4 As shown, the robot has 5 active degrees of freedom and 7 passive degrees of freedom, making it an overall 12-DOF intelligent robot. Furthermore, in local motion, since the wheels are omnidirectional wheels, each equipped with 12 small rollers, there are a total of 6 omnidirectional wheels, resulting in 72 local degrees of freedom. Regarding the robot's drive structure, the front and rear ball wheels are driven by the drive motors of the robot's first and fourth joint units, while the middle rollers are driven by the drive motors of the robot's second and third joint units via bevel gear transmission.

[0019] For example, see Figure 5 The robot can carry the following components: LiDAR: The basic model uses the Livox Mid-360, with a 360° horizontal field of view, a 58° vertical field of view, and a point cloud density of 120 points / ㎡; for oily / high humidity environments, a 1550nm infrared laser is selected; High-definition camera: Hikvision MV-CA050-20GM, 30fps frame rate, with LED supplemental lighting to ensure image quality in low-light environments; for oily / water mist environments, the high-definition camera lens is covered with a nano anti-fouling coating and integrates an adaptive polarization filtering algorithm, improving image contrast by more than 40% when the oil coverage is ≤30%; The LiDAR uses a 1550nm wavelength model, and the point cloud data missing rate is controlled within 8%. The sensor packaging uses polytetrafluoroethylene material, which can withstand environments with 95%RH (30℃) humidity and H2S concentration ≤500ppm. The circuit interface is an IP68 waterproof connector, which can maintain stable communication after immersion in 0.5m water for 24 hours.

[0020] Ultrasonic sensor: TIIWR6843, detection range 0.1-5m, sampling rate 500Hz, wall thickness measurement accuracy ±0.1mm, suitable for metal and non-metal pipe inspection; IMU + odometer: Huaxin WIS300 IMU zero bias instability 0.3° / h + high precision wheel encoder, supports inertial navigation in environments without GPS, positioning accuracy 0.1%.

[0021] Edge computing platform: NVIDIA Jetson Orin Super, 170 TOPS computing power, supporting real-time AI inference; Algorithm model: LOAM-SLAM algorithm to achieve centimeter-level positioning, YOLOv8 model to complete defect identification, LSTM behavior prediction model to predict obstacle trajectory.

[0022] Rotary actuator: Maxon EC-4pole22 motor, with a harmonic reducer and a reduction ratio of 1:50, output torque of 5Nm, meeting the motion requirements of complex pipelines; Linear actuator: Customized expansion mechanism, stroke 50-200mm, positioning accuracy ±0.1mm, supports multi-diameter DN150-DN300 adaptive; Dexterous actuator: Coreless motor + encoder, achieving millimeter-level operating accuracy.

[0023] Communication module: Huawei MH5000-314G / 5G dual-mode module, with a transmission rate of 100Mbps, supporting long-distance communication in field scenarios; Power system: Panasonic NCR18650 lithium battery, with a battery life of 4-8 hours, combined with electromagnetic induction wireless charging, to meet the needs of long-distance detection.

[0024] The dynamic energy consumption model is refined: power consumption in normal detection mode is ≤15W, with a battery life of 8 hours; instantaneous power consumption in obstacle crossing mode is ≤30W; in non-detection sections, it automatically switches to low-power sleep mode, shutting down the LiDAR and retaining only IMU + Bluetooth communication, reducing power consumption to 5W and extending battery life to 12 hours. It supports reserving a DC24V charging interface at the pipeline maintenance port, and can be equipped with a solar auxiliary module for exposed pipeline scenarios.

[0025] S102: Based on pose data and environmental data, construct a three-dimensional digital map of the pipeline and achieve real-time self-localization.

[0026] See Figure 6 After entering an unfamiliar pipeline lacking prior information, the robot uses its embedded computing platform to run a multi-sensor fusion synchronous localization and mapping algorithm in real time, so as to complete the three-dimensional modeling of the pipeline environment and its own precise positioning in parallel during the movement.

[0027] Specifically, the raw point cloud data collected by the LiDAR is first preprocessed locally on the robot. Voxel filtering and statistical filtering effectively remove noise and outliers, achieving a noise removal rate of over 95%. The processed point cloud data is then used for frame matching and pose estimation using an optimized LOAM-SLAM algorithm, incrementally constructing a continuous 3D point cloud map of the pipe's interior. This enables real-time mapping in unfamiliar environments with an accuracy of ±5 cm. Simultaneously, continuous displacement data from the wheeled odometry is tightly coupled and fused with stable feature points in the point cloud. An extended Kalman filter algorithm running on the robot itself achieves optimal estimation of multi-source sensor data, thereby calculating and updating the robot's position coordinates and trajectory in its self-constructed 3D point cloud map in real time. This ensures that the positioning error remains consistently less than a preset distance (e.g., 10 cm) within the confined space of the pipe, where there is no GPS signal and no external references. The triaxial acceleration and angular velocity information provided by the inertial measurement unit further assists in the calculation of the robot's motion attitude. Especially in complex structures such as pipe bends and reducers that cause sudden changes in motion state, the quaternion complementary filtering algorithm running on the main processor fuses the data from the IMU and wheel odometry in real time to output stable attitude information of the robot's roll, pitch and yaw angles, with an update frequency of no less than 100 Hz.

[0028] Furthermore, to maintain accuracy over long distances, a closed-loop detection and correction mechanism can be introduced. By matching the real-time scanned point cloud data with the constructed historical map fragments, the historical areas to be revisited can be intelligently identified, thereby effectively eliminating the error accumulated by the wheel odometer over time. The cumulative error of the entire detection trajectory can be controlled within 0.3%.

[0029] Based on this, the robot can generate a complete 3D digital map that integrates the geometry of the pipeline (including diameter, bending angle, connection parts, etc.), obstacle information and its own motion trajectory while operating completely autonomously. This map serves as the core of the robot's internal environment cognition, dynamically marking its real-time pose and detected areas, and providing an accurate spatial reference framework for subsequent defect identification, task planning and collaborative decision-making.

[0030] In addition, see Figure 7 To adapt to the complex working conditions that may be encountered inside the pipeline, the robot's wheelbase and posture can be dynamically adjusted via linear actuators to ensure that the sensors are always facing the pipe wall at the optimal distance, guaranteeing the quality of point cloud and image data acquisition. Even with pipe diameter changes of ±30%, stable mapping and positioning accuracy can still be maintained. When the LiDAR temporarily fails, it can automatically switch to a vision-and-IMU fusion positioning mode, using the ORB-SLAM3 algorithm to continue the positioning and mapping tasks. This ensures that the cumulative trajectory error is controlled within 0.8%, guaranteeing the continuity of the detection mission in harsh and unknown environments.

[0031] S103: Identify pipe wall defects in the pipeline based on pipe wall sensing data, generate corresponding defect data packages, and bind and store them with spatial location information in the three-dimensional digital map.

[0032] Based on the construction and real-time localization of the pipeline's 3D digital map, the robot utilizes its multiple onboard sensing modalities and pre-installed intelligent algorithms to accurately analyze the pipe wall condition and identify defects. High-definition cameras capture images of the pipeline's inner wall, which undergo adaptive image enhancement processing. These images are then analyzed in real-time by a pre-trained YOLOv8 target detection model deployed locally on the robot. This model identifies defects such as cracks, corrosion, and deformation on the pipe wall surface, achieving a detection accuracy of over 95% for microcracks smaller than 0.5 mm, with a classification confidence level of at least 0.95. An ultrasonic sensor detects the pipe wall thickness in real-time at a 500Hz sampling rate. By compensating for temperature drift using a BP neural network, the thickness measurement error is reduced to within ±0.1 mm, accurately identifying areas of wall thinning caused by corrosion or wear. Point cloud data provided by lidar further assists in the extraction of geometric features of defects. An analysis model based on these point cloud geometric features is specifically designed to identify morphological anomalies in the pipe wall. This analytical model constructs a quantitative description of the geometric morphology of the pipe wall surface by calculating the changes in point cloud density and local curvature distribution, thereby effectively identifying smooth deformation defects such as dents and bulges that are difficult to detect by traditional vision or ultrasonic sensing.

[0033] For each suspected defect identified, a local high-definition image of the defect area and the corresponding point cloud segment are automatically captured, and the abnormal section in the ultrasonic thickness curve is recorded simultaneously. By using the precise robot pose coordinates provided by the real-time self-localization module and the spatial mapping relationship of the three-dimensional digital map, the three-dimensional position information of the defect in the global coordinate system of the pipeline is calculated, and its coordinate error can be controlled within 10 centimeters.

[0034] See Figure 8 Specifically for metal pipes, eddy current sensors are used to re-inspect corrosion defects. By taking two measurements at the defect point, one forward and one backward, and comparing the data deviations, if the deviation is not greater than a set threshold, the average value is taken as the final result; otherwise, it is marked as data anomaly and both measurement values ​​are retained. This ensures the reliability and traceability of the data.

[0035] Subsequently, the aforementioned defect types, classification confidence levels, defect images, associated point cloud fragments, ultrasonic thickness curves, 3D location information, and eddy current retest data are integrated into a structured defect data package, which is then standardized and encapsulated using JSON format. Each generated defect data package is strongly correlated with the robot's trajectory points and corresponding spatial location information in the 3D digital map via timestamps and spatial coordinates, ensuring that each defect record can be accurately located and visualized backtracking within the pipeline model. The defect data packages are stored in real-time on the robot's local solid-state drive, and a data integrity check code is generated using the SHA-256 hash algorithm and stored together. This constructs a pipeline defect database at the data level that is accurately located, informationally complete, and tamper-proof, providing a solid data foundation for subsequent pipeline health assessments and maintenance decisions.

[0036] Furthermore, to further enhance the comprehensive judgment capability of defect identification, a fusion reasoning mechanism can be introduced during data processing. In the defect identification process, a factor graph optimization algorithm deeply fuses heterogeneous features from different sensors. The geometric features of point clouds, the texture features of images, and the material features of ultrasound are incorporated into an optimization model with more than 30 nodes for global consistency constraints and joint optimization. This ensures the unity and coordination between multi-sensor data at the feature level, controlling the reprojection error of visual features to within one pixel. Simultaneously, a built-in knowledge graph system is activated and participates in real-time reasoning. This knowledge graph system relies on a database containing more than 2000 historical defect cases, providing background knowledge and correlation reasoning support for the currently identified defect features. Through this case-based analogical reasoning, it not only relies on the instantaneous readings of the current sensors but also draws on historical experience. This further improves the accuracy and reliability of defect classification when facing complex situations with atypical morphology or severe interference, especially for rare defect types with scarce training samples, where the identification and judgment capabilities are significantly enhanced.

[0037] S104: When a neighboring robot is detected through the local communication module, a temporary communication link is established.

[0038] In this context, neighboring robots refer to other pipeline robots located within the robot's preset communication range. In the application, the robot continuously broadcasts its own identifier and scans surrounding signals at specific time intervals using its onboard local communication module, such as a ZigBee Mesh network or Bluetooth Low Energy, to detect the presence of other pipeline robots—i.e., neighboring robots—entering its preset communication range. Once a valid signal from a neighboring robot is detected and authentication is completed, the two automatically negotiate and establish a point-to-point temporary communication link. This link communicates through dynamically allocated temporary channels and features data encryption and collision avoidance mechanisms to ensure the security and reliability of information exchange in complex pipeline environments.

[0039] The establishment of this temporary link does not rely on any pre-deployed network infrastructure; it is entirely initiated and maintained autonomously by the robot, demonstrating its high degree of autonomous collaboration in unknown pipeline environments. Through this temporary communication link, the robot can conduct preliminary data interaction with neighboring robots, laying the physical foundation for subsequent exchange of detection information, including map summaries and task status information. This mechanism effectively breaks down the information silos of individual robots, providing technical support for achieving direct, peer-to-peer communication between multiple robots in closed pipelines lacking external networks, making information sharing and collaborative decision-making among groups possible.

[0040] S105: Exchange detection information with neighboring robots based on temporary communication links.

[0041] The detection information includes map summary data and task status information. The map summary data includes a set of key landmarks extracted from a 3D digital map and a pipeline topology diagram. The task status information includes runtime resource status information and task execution progress information.

[0042] Map summary data is a lightweight representation of environmental features extracted from the 3D digital map constructed by the robot itself. Its core content includes a set of key landmarks extracted from the map and a pipeline topology diagram representing pipeline connections. It aims to summarize the core spatial structure of the robot's explored environment with minimal data volume, adapting to the limited bandwidth of temporary links. Task status information is a dynamic description of the robot's operating status and task progress, specifically including operational resource status information (such as remaining battery power and computational load) and task execution progress information (such as detected pipeline areas and an overview of identified defects).

[0043] During the exchange, both robots standardize and encapsulate their respective detection information and transmit it to each other via a secure temporary link. This direct data exchange allows each robot to quickly learn about the map overview of the areas explored by its neighbors in the unknown environment and their current task load, providing crucial information input for subsequent map updates and task negotiation. Through this intermittent, lightweight information exchange, multiple robots can break down information barriers and initially form a shared understanding of the pipeline network environment. This lays a key data foundation for shifting from independent individual exploration to coordinated swarm intelligence operations, effectively avoiding redundant exploration and resource consumption in the vast pipeline network.

[0044] S106: Update the 3D digital map based on map summary data from neighboring robots.

[0045] After successfully receiving map summary data from a neighboring robot, the robot first performs spatial alignment processing to match its own 3D digital map with the received map summary data, determining the relative spatial transformation relationship between the two. This process is mainly achieved by comparing the set of key landmarks shared by both maps and the pipeline topology diagram, thereby accurately transforming the local map data of the neighboring robot into the global coordinate system of its own map.

[0046] Subsequently, based on the determined relative spatial transformation relationship, the newly added environmental features contained in the map summary data of neighboring robots, such as information on pipe branches, bends or obstacles that have not been explored by the robot itself, are integrated into its own three-dimensional digital map, thereby expanding the map coverage and supplementing and updating the content, effectively expanding the environmental cognition boundary of a single robot.

[0047] When, during the data fusion process, inconsistencies are found between the map summary data of neighboring robots and their own 3D digital maps in a certain overlapping area, the maps are corrected and updated according to a preset conflict resolution strategy. This strategy can make decisions based on the age of the data acquisition timestamps, the accuracy level of the sensor data, or pre-set confidence rules to ensure that the final generated map version has higher consistency and reliability.

[0048] Through this map update mechanism based on local communication, each robot can use collective perception information to continuously optimize and enrich its structured cognition of the environment, significantly improving the efficiency and accuracy of building a globally consistent environment model in complex pipeline networks, and providing a more solid and unified spatial context for subsequent collaborative path planning and task allocation.

[0049] S107: Based on the task status information of neighboring robots and the updated 3D digital map, negotiate tasks with neighboring robots.

[0050] In the application, the robot first comprehensively identifies pipeline areas that have not yet been fully detected based on the latest updated 3D digital map that incorporates swarm perception information, and generates a candidate task set accordingly. Subsequently, the robot combines its own task status information, including its operating resource status (such as remaining power and computing load) and task execution progress, to perform a cost assessment on each task in the candidate task set, calculating its own estimated cost for performing the task. This cost comprehensively considers multiple factors such as path length, energy consumption, time, and task priority.

[0051] After completing the calculations, the robot sends its estimated cost to neighboring robots via a temporary communication link, while simultaneously receiving neighboring robots' calculated estimated costs for the same candidate tasks. Based on this, the two robots conduct one or more rounds of bidding negotiation using all received cost information. This process is a distributed task allocation and optimization process, aiming to optimize overall detection efficiency at the robot swarm level, rather than just considering individual optimalities.

[0052] Based on the final agreed-upon bidding results, the robot updates its task queue, clarifying its next detection tasks. Through this collaborative task planning mechanism based on real-time communication and negotiation, multiple robots can work as an organic whole, dynamically adjusting their respective work priorities, thereby significantly improving the overall detection efficiency in complex pipeline networks, effectively avoiding redundant detection and resource waste, and achieving true autonomous collaboration and swarm intelligence.

[0053] S108: Adjust its target work area and movement path according to the task negotiation results.

[0054] After completing task negotiation with neighboring robots and updating its own task queue, the robot immediately adjusts its preset target work area dynamically according to the new task allocation scheme determined through negotiation. For example, based on the negotiation results, the robot may abandon a pipe section that was already efficiently covered by neighboring robots in the original plan, and instead focus its target work area on a pipe branch or specific segment that has not yet been detected and agreed upon by both parties. Based on this updated target work area and the latest 3D digital map that incorporates swarm perception information, the robot invokes its own path planning algorithm to recalculate and generate an optimal movement path from its current position to the new target area.

[0055] The path planning process fully considers the pipeline's topology, the distribution of known obstacles, and the robot's position relative to neighboring robots to ensure path feasibility and avoid conflicts during travel. The path adjustment is not static; as the robot travels along the new path and continues performing pipeline inspection tasks, it continuously perceives and localizes its environment. If it encounters new obstacles, or if environmental changes render the original path impassable, or if it encounters another robot and triggers a new negotiation, its path can be dynamically adjusted again.

[0056] Through this dynamic area allocation and path planning based on negotiation results, multiple robots achieve reasonable decomposition of tasks and coordinated actions in space. This enables the entire robot swarm to collaboratively cover and inspect the vast and complex pipeline network in a parallel, efficient, and non-repetitive manner, ultimately significantly improving the overall efficiency and completeness of pipeline inspection operations.

[0057] In one embodiment, each pipeline robot is also equipped with a wheeled odometer to collect mileage data; the above-mentioned construction of a three-dimensional digital map of the pipeline based on pose data and environmental data, and the realization of real-time self-localization, includes: It determines whether the current pipeline area is a historically traversed area; if so, it loads the corresponding historical 3D digital map as a priori map for self-localization, and integrates mileage data and environmental data, using a map matching localization algorithm to achieve self-localization and eliminate accumulated odometer errors; if not, it performs time synchronization and inter-frame matching on the pose data and current point cloud data in the environmental data to incrementally construct a new 3D digital map; by integrating the point cloud feature points in the mileage data and current point cloud data, it calculates and updates its own position coordinates and movement trajectory in the new 3D digital map in real time.

[0058] In the application, the robot possesses the ability to remember and reuse previously explored areas. The robot first intelligently assesses its current location within a pipeline, identifying whether it has previously traversed such areas. Specifically, this assessment deeply integrates real-time self-localization results with historical environmental data. As the robot continuously operates within its self-built and maintained 3D digital map, its localization algorithm calculates its precise pose in the map coordinate system by matching real-time sensor observations with the global map model. Based on this, a hierarchical decision-making process can be executed: First, if the real-time localized coordinates clearly fall within the boundary of any existing historical map area, it is directly determined that the robot is currently in a previously traversed area. Second, when the robot approaches or is in a fuzzy boundary zone of a known map, increasing localization uncertainty, a closed-loop decision-making process based on perceptual features is activated: key environmental features from the current sensor data are extracted and quickly compared with the feature database of the historical map. For example, the robot can extract key features (such as specific pipe wall textures, structural welds, flange connections, or local curvature features) from the current LiDAR point cloud or visual image, and quickly compare them with the feature database of the corresponding recorded areas in its stored historical 3D digital map. If the matching degree exceeds a preset threshold, it is determined that it has entered a previously traversed area. If neither of the above two conditions is met, i.e., the positioning coordinates are not in any known historical area, and the real-time perceived features cannot be effectively matched with any historical area, it can be determined that it has entered a completely new, unexplored pipeline area, and an incremental map building process for the new area is immediately triggered.

[0059] If the area is determined to be a previously visited area, the robot loads the corresponding historical 3D digital map from its local storage as a priori map for self-localization. Based on this, by fusing real-time wheel odometer data with point cloud features in the current environmental data, a map matching localization algorithm is used to closely match real-time sensor observations with the priori map, thereby achieving precise calibration of its own position. This process can effectively eliminate the cumulative error that inevitably occurs in wheel odometers during long-distance operation and keep the positioning drift at a low level.

[0060] If the robot determines that the current area is unexplored, it initiates an incremental map-building process. This involves time synchronization and fine-grained inter-frame matching of the real-time pose data and the current point cloud data in the environment, gradually expanding its environmental model to construct a new 3D digital map. During this process, by continuously fusing continuous displacement information from the wheeled odometer with stable feature points extracted from the current point cloud data, the robot can calculate and accurately update its position coordinates and trajectory in the newly constructed 3D digital map in real time. This mechanism allows the robot to flexibly switch between "relocalization" and "map creation" modes in complex pipeline networks, ensuring both high-precision positioning in known areas and exploration and mapping capabilities in unknown areas. This significantly improves the overall positioning efficiency and accuracy in long-distance, repetitive pipeline inspection tasks. Figure 1 To the point of being responsive.

[0061] In one embodiment, the above-mentioned identification of pipe wall defects in a pipeline based on pipe wall sensing data and generation of corresponding defect data packets includes: Based on pipe wall sensing data, a pre-built few-sample learning model is used to identify pipe wall defects and determine the defect type and classification confidence of the pipe wall defects; defect images corresponding to the pipe wall defects are extracted from the pipe wall sensing data; based on the robot pose coordinates determined by real-time self-localization, the three-dimensional position information of the pipe wall defects in the pipeline is determined; and the defect type, classification confidence, defect image and three-dimensional position information are integrated into a defect data package.

[0062] In the application, the process of identifying pipe wall defects in the pipeline based on pipe wall sensing data and generating corresponding defect data packets is implemented by an integrated intelligent analysis system. The architecture of this intelligent analysis system is a few-shot learning model based on multi-source data fusion. Specifically, this few-shot learning model is not a single algorithm, but a collection of collaborative models that integrates specialized sub-models for different sensing modalities: For image data acquired by high-definition cameras, a pre-trained YOLOv8 object detection model is used as its visual analysis engine to identify and locate defects such as cracks and corrosion on the pipe wall surface from complex backgrounds, and to provide preliminary type judgments and classification confidence levels; For thickness data acquired by ultrasonic sensors, a dedicated BP neural network is embedded to process the raw signal, accurately calculate the wall thickness, and compensate for drift caused by factors such as temperature, thereby identifying wall thickness reduction defects; For 3D point cloud data provided by lidar, an analysis model based on point cloud geometric features is integrated. This analysis model constructs a quantitative description of the geometric morphology of the pipe wall surface by calculating the changes in point cloud density and local curvature distribution, thus specifically used to identify smooth deformation defects such as dents and bulges in the pipe wall that are difficult to detect visually and ultrasonically.

[0063] Simultaneously, the real-time pose coordinates provided by the positioning module are invoked. Through coordinate transformation and map mapping, the three-dimensional location information of the defect in the pipeline's global coordinate system is accurately calculated, with the positioning error controlled within the centimeter level. Subsequently, the aforementioned multi-source information—defect type, classification confidence level, defect image, and three-dimensional location information—is standardized and encapsulated, integrated into a structured defect data package. This data package can be organized using a lightweight data format (such as JSON) for easy storage and subsequent transmission.

[0064] In one embodiment, updating the 3D digital map based on map summary data from neighboring robots includes: The relative spatial transformation relationship between the 3D digital map and the map summary data of adjacent robots is determined through spatial alignment processing. Based on the relative spatial transformation relationship, the newly added environmental features in the map summary data of adjacent robots are integrated into the 3D digital map to complete the expansion and update of the 3D digital map. When the map summary data of neighboring robots and the 3D digital map have inconsistent representations in overlapping areas, the 3D digital map is corrected and updated according to a preset conflict resolution strategy.

[0065] In this application, the process of a robot updating its own 3D digital map is achieved through a systematic data fusion and conflict resolution mechanism. This process begins with spatial alignment, where efficient point cloud registration algorithms (such as the Iterative Closest Point algorithm and its variants) match the key landmark sets and pipeline topology in both robots' maps. This accurately calculates the relative spatial transformation between the robot's own 3D digital map and the received map summary data, establishing a unified coordinate reference for subsequent data fusion. After establishing the spatial alignment, the map content is then fused and updated. New environmental features not yet recorded in the robot's own map, such as unexplored pipeline branches, bends, or obstacles, are identified in the neighboring robot's map summary data. Based on the aforementioned relative spatial transformation, these new features are accurately integrated into the robot's own 3D digital map, effectively expanding the map's coverage and seamlessly enriching its content.

[0066] Crucially, if inconsistencies are detected in the environmental representations of the same overlapping area by both parties during the fusion process (e.g., different descriptions of the location or shape of the same obstacle, contradictory records of the connection relationships of pipe branch structures, or inconsistent pipe diameter measurement data for the same area), the system will automatically adjudicate and correct the issue based on a preset conflict resolution strategy. This strategy can be based on various rational principles, such as prioritizing map information with more recent data acquisition time, trusting data collected by higher precision sensors, or selecting based on the confidence level preset for different robots, thereby covering or correcting questionable parts in its own 3D digital map.

[0067] In one embodiment, the above-mentioned task negotiation with neighboring robots based on the task status information of neighboring robots and the updated 3D digital map includes: Based on the updated 3D digital map, identify the area to be detected and generate a candidate task set; based on the task status information, calculate the estimated cost of executing each task in the candidate task set; send the estimated cost to neighboring robots and receive the estimated costs from neighboring robots; based on the estimated cost and the neighboring costs, conduct one or more rounds of bidding negotiation with neighboring robots; and update the task queue according to the bidding negotiation results.

[0068] In the application, the robot first intelligently identifies all undetected pipeline areas based on a state-of-the-art 3D digital map incorporating swarm perception information, formalizing them into a specific set of candidate tasks. Then, the robot deeply analyzes its own task status information, including its operational resource status (such as remaining power and computing load) and current task execution progress. Based on this, it calculates its estimated cost for each task in the candidate task set. This cost is a quantitative indicator that comprehensively considers multiple factors such as path length, energy consumption, estimated time, and task priority, aiming to objectively reflect the degree of resource consumption of executing the task. After calculation, the robot sends its estimated cost to neighboring robots via existing temporary communication links, while simultaneously receiving neighboring estimated costs for the same candidate tasks calculated based on their own status.

[0069] Based on this, the two robots engage in one or more rounds of bidding negotiations using all cost information. This process resembles a distributed contract network protocol, aiming to find the optimal task allocation scheme from a system-wide perspective, rather than merely pursuing individual optimality, through information exchange and repeated game theory. Ultimately, based on the agreed-upon bidding results, the robots dynamically update their task queues, explicitly including newly assigned tasks and potentially releasing existing tasks that have been more efficiently handled by other robots. This cost-comparison-based distributed negotiation mechanism enables the robot swarm to autonomously and efficiently achieve dynamic allocation and load balancing of detection tasks without relying on a central controller. This significantly improves the overall efficiency of complex pipeline network detection operations at the system level and effectively avoids work duplication and resource waste.

[0070] Furthermore, if task negotiation with neighboring robots fails, a distributed fault-tolerant processing procedure is triggered, including at least one of the following: The system implements several mechanisms: a negotiation retry mechanism, a negotiation mechanism, and a task re-initiation for the same candidate task set after a preset delay time, adjusting negotiation parameters during the retry; an autonomous decision-making mechanism, which ignores negotiation failures and autonomously adds the failed task to its own task queue when at least one task in the candidate task set has a priority higher than a preset threshold; a virtual marking mechanism, which adds virtual conflict markers to the spatial locations of the failed candidate task sets in the 3D digital map, synchronizing these markers as part of the map summary data during subsequent data exchange with other robots; and a task shelving mechanism, which re-marks failed candidate tasks as pending assignments and lowers their priority in the local task queue.

[0071] In applications, to address scenarios where task negotiation with neighboring robots fails, multiple optional autonomous decision-making strategies can be set, allowing the robot to choose to execute one or more of them based on its own state and task context.

[0072] One approach is to implement a negotiation retry mechanism. This involves the robot re-initiating task negotiation for the same set of candidate tasks after a preset delay (e.g., several seconds). During the retry, the robot adjusts the negotiation parameters to increase the probability of successful negotiation. The negotiation parameters are an adjustable set of strategies, which may include: adjusting the weighting coefficients in its estimated cost calculation model (e.g., increasing or decreasing the weight given to path length and energy consumption), introducing a random perturbation value to the cost bid to break the negotiation deadlock, or adjusting the upper limit of the bidding rounds and the time interval between each bid.

[0073] Secondly, it implements an autonomous decision-making mechanism. When the robot determines through built-in rules that the priority of one or more candidate tasks is higher than a preset threshold (such as involving the re-inspection of serious defects or the exploration of emergency areas), it will ignore the result of the failed negotiation, autonomously add the high-priority task to its own task queue and force its execution, ensuring that critical tasks are not delayed.

[0074] Thirdly, a virtual marking mechanism is implemented. The robot adds a special virtual conflict marker to the spatial location of the candidate task that failed in this negotiation in its 3D digital map. This marker will be broadcast as part of the map summary data when synchronizing maps with other robots in the future, thereby alerting other members of the group that there was a task assignment conflict in the area, which can be used by other robots for reference and careful handling.

[0075] Fourthly, there is a task shelving mechanism, which simply re-marks the candidate tasks that failed to negotiate as "pending assignment" and temporarily lowers their priority in the local task queue, waiting for them to be proposed again in subsequent cycles or to be claimed by other robots.

[0076] This multi-layered, distributed fault-tolerance mechanism ensures that the temporary failure of one or more negotiation nodes will not cause the entire collaborative inspection task to stagnate or become chaotic, thus guaranteeing the final completion rate of long-cycle, large-scale pipeline inspection tasks.

[0077] In one embodiment, each pipeline robot is also equipped with a delivery device, and the above detection method further includes: When a pipeline direction is selected at a pipeline branch node, the beacon deployment device is controlled to deploy a physical beacon on the inlet pipe wall of the selected pipeline direction. During pipeline exploration, a multimodal sensor is used to identify whether there are physical beacons deployed by other pipeline robots on the inlet pipe walls of multiple selectable pipeline directions. Based on the physical beacon identification results, a pipeline direction that has not been deployed with physical beacons by other pipeline robots is selected as the target exploration direction.

[0078] In applications, when a robot makes a path decision at a branch node in a pipeline, it controls a delivery device to deploy a physical beacon with a specific identifier on the inlet pipe wall of the pipeline direction it ultimately selects. This beacon can be a visual tag with a unique pattern or color, a magnetic tag that can be identified by a magnetic sensor, or an RFID tag with an embedded near-field communication chip, etc., which is intended to provide clear physical instructions for subsequent passing robots.

[0079] Subsequently, when any robot reaches a pipe branch node and makes an exploration decision among multiple optional pipe directions, it proactively scans and identifies whether physical beacons placed by other robots exist on the walls of each inlet pipe using its multimodal sensors (such as high-definition cameras for image recognition of visual tags, or magnetic sensors for detecting magnetic markers). Based on this identification result, the robot can execute a simple cooperative rule: prioritize selecting a pipe direction on its inlet pipe wall where no physical beacon has been placed as the target exploration direction. This mechanism effectively transmits the basic information "this path has already been explored" among the robot swarm in a low-cost, low-communication-overhead physical way, thereby greatly reducing the probability of multiple robots repeatedly exploring the same pipe branch at the system level, and significantly improving the overall exploration efficiency and coverage of the swarm in the vast pipe network.

[0080] Understandably, a robot might encounter a situation where all possible directions at a pipe branch node have already had physical beacons deployed by other robots on the inlet pipe walls. To address this specific scenario and ensure the continuous progress of the detection mission, the robot will no longer apply the default rule of "prioritizing directions without beacons," but will instead initiate an auxiliary decision-making process based on additional information.

[0081] This decision support process can be based on one or more of the following strategies: First, the robot can attempt to read the additional information carried by the beacon (such as reading the deployment timestamp or robot identifier recorded in the RFID tag through near-field communication) and prioritize exploring the direction with the longest beacon deployment time; Second, the robot can combine its own task status information (such as remaining battery power and current task queue) with the limited global information obtained through wireless communication (such as the detection completion rate of certain areas learned from neighboring robots) to choose the direction with the lowest overall cost or the highest expected benefit for itself; Third, the robot can execute a random selection strategy to ensure that the group behavior can still maintain the diversity of exploration on a macro level and avoid all robots getting stuck in collective waiting at the same node.

[0082] This physical beacon mechanism can effectively complement the electronic map synchronization and task negotiation mechanism based on wireless communication. In scenarios with poor communication signals or low robot density, it can still maintain basic collaborative efficiency as a degraded collaboration strategy, thereby enhancing the environmental adaptability and robustness of the entire multi-robot system under different working conditions.

[0083] In one embodiment, the above detection method further includes: The system monitors its remaining power in real time; based on historical energy consumption data and a 3D digital map, it dynamically predicts the minimum return energy required to return to the pipeline inlet or communication section; the communication section is the section that can communicate with the outside world of the pipeline; it compares the remaining power with the minimum return energy; if the remaining power is lower than or equal to the sum of the minimum return energy and a preset safety redundancy, it performs a return operation.

[0084] During pipeline inspection operations, the robot can also calculate its energy consumption status in real time using a dynamic energy consumption model. This model continuously monitors the remaining battery power in the robot's power system with an accuracy of ±1%. Simultaneously, this model comprehensively considers geometric information provided by a 3D digital map, such as pipeline slope and turning frequency, as well as historical energy consumption data, to dynamically calculate the minimum return energy required for the robot to return from its current position to the pipeline entrance or a communication point where stable communication can be established. Based on this, the real-time remaining battery power is continuously compared with the sum of the minimum return energy consumption and a preset safety redundancy. This safety redundancy is designed to address the uncertainties caused by unexpected situations within the pipeline (such as encountering unexpected obstacles requiring additional energy to overcome them), serving as a crucial buffer to ensure the reliability of decision-making. Once the remaining battery power is detected to be below or equal to this comprehensive threshold, a return command is immediately generated, controlling the robot to automatically terminate the current inspection task and initiate the return procedure.

[0085] During the return journey, the robot retreats along its original trajectory or the optimal energy-saving path planned by the 3D digital map. The trajectory is corrected in real time by comparing pipe wall feature data from the round trip phase, keeping the return error within 1%. The primary goal is to ensure the safety of the collected core data. To maximize endurance, the robot automatically switches to a power-saving operation mode during this phase: unnecessary functions such as high-definition camera illumination, LiDAR mapping, and continuous ultrasonic thickness detection are disabled, maintaining only basic positioning, environmental perception, and core communication, and simplifying data uploads. All data related to energy consumption prediction, decision threshold comparison, and return execution are recorded in the system log in real time, providing data support for subsequent task planning and battery performance analysis.

[0086] This intelligent autonomous decision-making based on accurate energy consumption prediction and safety redundancy can effectively avoid the risks of robots getting stuck, mission interruption, or loss of valuable inspection data due to power depletion, significantly improving the reliability and continuity of long-distance pipeline inspection tasks.

[0087] In one embodiment, the above detection method further includes: The robot can identify obstacles in the environmental data in real time and determine whether they are passable. If they are not passable, the robot will mark the obstacle information and control the robot to perform an obstacle avoidance and retreat operation. If they are passable, the robot will adjust its motion posture and continue to run.

[0088] For example, the above-mentioned judgment on passability may specifically be: based on a three-dimensional digital map, determining whether the height of the obstacle is less than 1 / 3 of the pipe diameter, and / or determining whether the obstacle type is deformable sediment.

[0089] See Figure 9 During pipeline inspection, the robot operates in an unknown pipeline environment, using multimodal sensors to perceive the surroundings in real time and intelligently identify and autonomously determine the passability of detected obstacles. The lidar scans the area in front of the pipeline at a 10Hz frequency, generating high-density point cloud data. Simultaneously, it analyzes the real-time images using the YOLOv8 visual recognition algorithm to accurately identify obstacle types, including different categories such as stone deposits, structural deformations of the pipe wall, and protruding foreign objects, and precisely measures their geometric dimensions based on the point cloud data. Based on this, the robot autonomously and comprehensively judges the passability of obstacles using its real-time maintained 3D digital map and perception data. Its core judgment criteria include: whether the maximum height of the obstacle does not exceed one-third of the current pipeline inner diameter; whether the obstacle's physical type is deformable deposits (such as soft silt); and whether the obstacle may cause a structural collapse risk based on point cloud morphology analysis.

[0090] When an obstacle is determined to be impassable, the robot can attempt to broadcast obstacle alarms and precise location information to the vicinity of potential pipeline access points via its onboard local communication module (such as a ZigBeeMesh network). Its positioning relies on real-time self-localization results. Simultaneously, the robot automatically records and stores detailed information about the obstacle, including its 3D coordinates, multi-angle images, point cloud data, and dimensional parameters, in its local solid-state storage, forming a complete obstacle profile. Subsequently, it performs an obstacle avoidance and retreat operation, reversing at a constant speed along its predetermined trajectory. During the retreat, it continuously perceives the environment and collects data, comparing it with data from the forward phase to ensure the safety of the retreat path. The collected obstacle profile provides crucial decision-making information for subsequent task planning or human intervention.

[0091] When an obstacle is determined to be passable, the robot dynamically adjusts its motion strategy and control parameters based on the identified obstacle type. For obstacles such as pipe wall deformation, the robot reduces its speed to 0.1 m / s and uses its onboard rotary actuator to perform a 360-degree surround shot, comprehensively recording the condition of the deformed area. For obstacles such as protruding objects, a smooth motion control algorithm fine-tunes the robot's yaw angle, adjusting its posture within a range of ±15 degrees to find a safe passage and avoid collisions. Throughout the process of handling all passable obstacles, the robot updates and accurately marks the location and characteristic attributes of obstacles in its 3D digital map in real time, while maintaining the continuity of defect detection operations as much as possible.

[0092] Throughout the obstacle handling process, the robot's communication module maintains its established communication strategy, while the power management system monitors the energy consumption status in real time. If it enters the obstacle crossing operation mode, it automatically activates the power enhancement mode to provide additional torque, while limiting the instantaneous power consumption to a safe threshold (such as 30 watts).

[0093] In one embodiment, after the embodied intelligent pipeline robot exits the pipeline, the above detection method further includes: Upload the bound defect data package and 3D digital map to the digital twin system; The severity of pipe wall defects is rated based on historical defect data, and an operation and maintenance recommendation report is generated.

[0094] See Figure 10 , Figure 11 and Figure 12 In the application, after the embodied intelligent pipeline robot completes the pipeline inspection task and exits the pipeline, it automatically initiates the data upload and post-processing process. The robot first transmits the locally stored defect data package and 3D digital map to the cloud-based digital twin system via the communication module in an encrypted manner. The transmission process uses the AES-256 encryption algorithm to ensure data security, and at the same time, it uses the SHA-256 hash algorithm to verify data integrity, ensuring that no data is lost or tampered with during transmission.

[0095] After receiving data, the digital twin system automatically performs multi-source data fusion and correlation analysis. It calls upon a pre-trained domain-adaptive large model to perform in-depth processing on the uploaded defect data: First, based on over 2000 case samples in the historical defect database, it uses knowledge graph reasoning technology to rate the severity of each defect, classifying them into levels I-IV. The rating criteria comprehensively consider multiple factors such as defect type, size, wall thickness reduction rate, location, and development trend. Then, it matches a knowledge base containing over 500 historical repair solutions to generate personalized operation and maintenance suggestions for each defect, including specific parameters such as repair process, material requirements, construction cycle, and re-inspection interval. The recommendation accuracy rate can reach over 85%.

[0096] The system automatically generates a comprehensive report containing a 3D visualization of defect distribution, a severity rating list, and detailed maintenance recommendations, supporting both PDF and Excel output formats. The report includes precise location information for each defect (X / Y / Z coordinates, error less than 10 cm), multi-angle defect images, ultrasonic thickness curve changes, and repair priority ranking, while also annotating data acquisition time and environmental parameters. Through a cloud-based GIS system and 3D visualization platform, maintenance personnel can remotely view the overall pipeline layout and defect distribution, and obtain on-site repair guidance via AR devices.

[0097] All uploaded data is synchronously used for model optimization and knowledge iteration. Through federated learning technology, while protecting data privacy, the edge-end small-sample learning model is optimized by combining real-time data collected by multiple robots, resulting in a 2% monthly improvement in defect identification accuracy. The optimized model parameters are then distributed to the robots via OTA (Over-The-Air) wireless upgrades, enabling continuous evolution of the algorithm's capabilities. The digital twin system synchronously updates the pipeline health status database, establishing a closed-loop management system covering the entire lifecycle from defect identification to maintenance and repair, providing data support and decision-making basis for preventative maintenance of the pipeline network.

[0098] Corresponding to the aforementioned application function implementation method embodiments, the present invention also provides a pipeline inspection system based on an embodied intelligent pipeline robot and corresponding embodiments.

[0099] Please see Figure 13 , Figure 13 This is a schematic diagram of the modular structure of a pipeline inspection system based on an embodied intelligent pipeline robot.

[0100] There are multiple pipeline robots, each equipped with multimodal sensors and local communication modules; the pipeline inspection system is operated by each pipeline robot, including: The data acquisition unit 131 is used to acquire pose data collected by the multimodal sensor, environmental data inside the pipeline, and pipe wall sensing data. The map building and positioning unit 132 is used to build a three-dimensional digital map of the pipeline based on pose data and environmental data and to achieve real-time self-positioning. Defect identification unit 133 is used to identify pipe wall defects in the pipeline based on pipe wall sensing data, generate corresponding defect data packets and bind and store them with spatial location information in the three-dimensional digital map; The communication management unit 134 is used to establish a temporary communication link when a neighboring robot is detected through the local communication module; the neighboring robot is another pipeline robot located within its preset communication range. The data exchange unit 135 is used to exchange detection information with neighboring robots based on a temporary communication link; the detection information includes map summary data and task status information. Map update unit 136 is used to update the three-dimensional digital map based on map summary data of neighboring robots; The task negotiation unit 137 is used to negotiate tasks with neighboring robots based on the task status information of neighboring robots and the updated three-dimensional digital map. The path planning unit 138 is used to adjust its target work area and movement path according to the task negotiation results.

[0101] In one embodiment, each pipeline robot is also equipped with a wheeled odometer to collect mileage data; regarding the construction of a three-dimensional digital map of the pipeline based on pose data and environmental data and the realization of real-time self-localization, the map construction and localization unit 132 is specifically used for: Determine whether the current pipeline area is a previously traversed area; If so, the corresponding historical 3D digital map is loaded as a priori map for self-localization, and mileage data and environmental data are integrated. A map matching positioning algorithm is used to achieve self-localization in order to eliminate the accumulated error of the odometer. If not, then time synchronization and inter-frame matching are performed on the current point cloud data in the pose data and environmental data to incrementally build a new 3D digital map. By fusing mileage data with point cloud feature points in the current point cloud data, it calculates and updates its position coordinates and movement trajectory in the new 3D digital map in real time.

[0102] In one embodiment, in identifying pipe wall defects in a pipeline based on pipe wall sensing data and generating corresponding defect data packets, the defect identification unit 133 is specifically used for: Based on pipe wall sensing data, a pre-built few-sample learning model is used to identify pipe wall defects and determine the defect type and classification confidence of the pipe wall defects; defect images corresponding to the pipe wall defects are extracted from the pipe wall sensing data; based on the robot pose coordinates determined by real-time self-localization, the three-dimensional position information of the pipe wall defects in the pipeline is determined; and the defect type, classification confidence, defect image and three-dimensional position information are integrated into a defect data package.

[0103] In one embodiment, regarding updating a 3D digital map based on map summary data from neighboring robots, the map updating unit 136 is specifically used for: The relative spatial transformation relationship between the 3D digital map and the map summary data of adjacent robots is determined through spatial alignment processing. Based on the relative spatial transformation relationship, the newly added environmental features in the map summary data of adjacent robots are integrated into the 3D digital map to complete the expansion and update of the 3D digital map. When the map summary data of neighboring robots and the 3D digital map have inconsistent representations in overlapping areas, the 3D digital map is corrected and updated according to a preset conflict resolution strategy.

[0104] In one embodiment, the task negotiation unit 137 is specifically used for: negotiating tasks with neighboring robots based on the task status information of neighboring robots and the updated 3D digital map; Based on the updated 3D digital map, identify the area to be detected and generate a candidate task set; based on the task status information, calculate the estimated cost of executing each task in the candidate task set; send the estimated cost to neighboring robots and receive the estimated costs from neighboring robots; based on the estimated cost and the neighboring costs, conduct one or more rounds of bidding negotiation with neighboring robots; and update the task queue according to the bidding negotiation results.

[0105] In one embodiment, each pipeline robot is also equipped with a delivery device, and the aforementioned detection system further includes: The beacon deployment control unit is used to control the beacon deployment device to deploy a physical beacon on the inlet wall of the selected pipeline direction when a pipeline direction is selected at a pipeline branch node. The beacon recognition unit is used to identify, through multimodal sensors, whether physical beacons placed by other pipeline robots are present on the inlet pipe wall in multiple selectable pipeline directions during pipeline exploration; The aforementioned path planning unit 138 is also used to prioritize selecting a pipeline direction that has not been deployed by other pipeline robots as the target exploration direction based on the physical beacon recognition results.

[0106] In one embodiment, the detection system further includes: a return-to-home control unit, configured to: The system monitors its remaining power in real time; based on historical energy consumption data and a 3D digital map, it dynamically predicts the minimum return energy required to return to the pipeline inlet or communication section; the communication section is the section that can communicate with the outside world of the pipeline; it compares the remaining power with the minimum return energy; if the remaining power is lower than or equal to the sum of the minimum return energy and a preset safety redundancy, it performs a return operation.

[0107] In one embodiment, the detection system further includes: an obstacle avoidance control unit, configured to: The robot can identify obstacles in the environmental data in real time and determine whether they are passable. If they are not passable, the robot will mark the obstacle information and control the robot to perform an obstacle avoidance and retreat operation. If they are passable, the robot will adjust its motion posture and continue to run.

[0108] In one embodiment, after the embodied intelligent pipeline robot exits the pipeline, the above-mentioned detection system further includes: an operation and maintenance decision unit, used for: Upload the bound defect data package and 3D digital map to the digital twin system; rate the severity of pipe wall defects based on historical defect data, and generate an operation and maintenance recommendation report.

[0109] Regarding the system in the above embodiments, the specific manner in which each unit performs operations has been described in detail in the embodiments related to the method, and will not be elaborated further here.

Claims

1. A pipeline inspection method based on an embodied intelligent pipeline robot, characterized in that, The pipeline robots are multiple in number, and each is equipped with a multimodal sensor and a local communication module; the method is executed by each of the pipeline robots, including: Acquire the pose data, internal environmental data of the pipeline, and pipe wall sensing data collected by the multimodal sensor; Based on the pose data and environmental data, a three-dimensional digital map of the pipeline is constructed and real-time self-localization is achieved; Based on the pipe wall sensing data, pipe wall defects in the pipeline are identified, corresponding defect data packages are generated, and they are bound and stored with the spatial location information in the three-dimensional digital map. When a neighboring robot is detected through the local communication module, a temporary communication link is established; the neighboring robot is another pipeline robot located within its preset communication range. Based on the temporary communication link, they exchange their respective detection information with the neighboring robots; the detection information includes map summary data and task status information. The 3D digital map is updated based on the map summary data of the neighboring robots; Based on the task status information of the neighboring robots and the updated 3D digital map, conduct task negotiation with the neighboring robots; Based on the results of the task negotiation, adjust your target work area and movement path.

2. The pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, Each of the pipeline robots is also equipped with a wheeled odometer to collect mileage data; based on the pose data and environmental data, a three-dimensional digital map of the pipeline is constructed and real-time self-localization is achieved, including: Determine whether the current pipeline area is a previously traversed area; If so, the corresponding historical 3D digital map is loaded as a priori map for self-localization, and the mileage data and environmental data are fused together. A map matching and positioning algorithm is used to achieve self-localization in order to eliminate the accumulated error of the odometer. If not, the pose data and the current point cloud data in the environment data are synchronized in time and matched between frames to incrementally construct a new 3D digital map. By fusing the mileage data with the point cloud feature points in the current point cloud data, the system calculates and updates its position coordinates and movement trajectory in the new three-dimensional digital map in real time.

3. The pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, Based on the pipe wall sensing data, pipe wall defects in the pipeline are identified, and corresponding defect data packets are generated, including: Based on the pipe wall sensing data, a pre-built few-shot learning model is used to identify pipe wall defects and determine the defect type and classification confidence level of the pipe wall defects. Extract the defect image corresponding to the pipe wall defect from the pipe wall sensing data; Based on the robot pose coordinates determined by real-time self-localization, the three-dimensional position information of the pipe wall defect in the pipeline is determined. The defect type, classification confidence level, defect image, and 3D location information are integrated into a defect data package.

4. The pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, Based on the map summary data of the neighboring robots, the 3D digital map is updated, including: The relative spatial transformation relationship between the 3D digital map and the map summary data of the adjacent robot is determined through spatial alignment processing. Based on the relative spatial transformation relationship, the newly added environmental features in the map summary data of the adjacent robots are fused into the three-dimensional digital map to complete the expansion and update of the three-dimensional digital map. When the map summary data of the adjacent robot and the three-dimensional digital map have inconsistent representations in the overlapping area, the three-dimensional digital map is corrected and updated according to a preset conflict resolution strategy.

5. A pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, Based on the task status information of the neighboring robots and the updated 3D digital map, task negotiation is conducted with the neighboring robots, including: Based on the updated 3D digital map, the area to be detected is identified and a set of candidate tasks is generated; Based on the task status information, calculate the estimated cost of executing each task in the candidate task set. Send its own estimated cost to the neighboring robot, and receive the neighboring estimated cost sent by the neighboring robot; Based on its own estimated cost and the estimated costs of neighboring robots, it conducts one or more rounds of bidding negotiations with the neighboring robots. Update its own task queue based on the bidding negotiation results.

6. A pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, Each of the pipeline robots is also equipped with a delivery device, and the method further includes: When a pipeline direction is selected at a pipeline branch node, the beacon delivery device is controlled to deliver a physical beacon on the inlet wall of the selected pipeline direction; During pipeline exploration, the multimodal sensor identifies whether there are physical beacons deployed by other pipeline robots on the inlet pipe wall in multiple selectable pipeline directions; Based on the physical beacon recognition results, a pipeline direction that has not been deployed by other pipeline robots is selected as the target exploration direction.

7. A pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, The method further includes: Monitor your remaining battery level in real time; Based on historical energy consumption data and the three-dimensional digital map, the minimum return energy required to return to the pipeline inlet or communication section is dynamically predicted; the communication section is a section that can communicate with the outside world of the pipeline. Compare the remaining power with the minimum return energy consumption; If the remaining power is less than or equal to the sum of the minimum return energy consumption and a preset safety redundancy, then a return-to-home operation is performed.

8. A pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, The map summary data includes: a set of key landmarks extracted from the three-dimensional digital map and a pipeline topology diagram.

9. A pipeline inspection method based on an embodied intelligent pipeline robot according to claim 1, characterized in that, The task status information includes runtime resource status information and task execution progress information.

10. A pipeline inspection system based on an embodied intelligent pipeline robot, characterized in that, The pipeline robots are numerous and each is equipped with a multimodal sensor and a local communication module; the detection system is carried and operated by each of the pipeline robots and includes: The data acquisition unit is used to acquire the pose data, internal environmental data of the pipeline, and pipe wall sensing data collected by the multimodal sensor. The map building and positioning unit is used to build a three-dimensional digital map of the pipeline based on the pose data and environmental data and to achieve real-time self-positioning. The defect identification unit is used to identify pipe wall defects in the pipeline based on the pipe wall sensing data, generate corresponding defect data packages, and bind and store them with the spatial location information in the three-dimensional digital map. The communication management unit is used to establish a temporary communication link when a neighboring robot is detected through the local communication module; the neighboring robot is another pipeline robot located within its preset communication range. A data exchange unit is used to exchange detection information with the adjacent robot based on the temporary communication link; the detection information includes map summary data and task status information. A map update unit is used to update the three-dimensional digital map based on the map summary data of the neighboring robots; The task negotiation unit is used to negotiate tasks with the neighboring robots based on the task status information of the neighboring robots and the updated three-dimensional digital map. The path planning unit is used to adjust its target work area and movement path based on the task negotiation results.

Images