A method and device for trenchless repair of urban underground sewer pipes

Abstract

This invention discloses a trenchless repair method and apparatus for urban underground drainage pipelines, comprising: acquiring 3D point cloud data of the pipeline interior structured light, LiDAR cross-sectional data, and distributed fiber optic acoustic sensing data collected by a robot, and transmitting them to a digital twin server; the digital twin server performs point cloud filtering and noise reduction processing on the collected data, directly identifying defect information and determining the 3D spatial location of the defect through the geometric features of the defect area in the point cloud; the digital twin server optimizes the repair process parameters by calling a knowledge graph and a repair effect prediction model based on the defect information and pipeline environmental parameters; the robot performs the repair operation according to the repair process parameters and provides real-time feedback on the quality data of the repair layer; the digital twin server compares the quality data with the expected indicators in the digital twin model, and adjusts the repair parameters and controls the robot to continue operation when the deviation exceeds a preset threshold.

Description

Technical Field

[0001] This invention relates to, in particular, a method and apparatus for trenchless repair of urban underground drainage pipes. Background Technology

[0002] Urban underground drainage pipes are a vital infrastructure component of modern cities, responsible for the transportation of sewage and the discharge of rainwater. Their operational status directly impacts urban water quality and the safety of residents. Statistics show that the total length of urban underground drainage pipes in my country exceeds 800,000 kilometers, a significant proportion of which have been in service for over 20 years. These pipes commonly exhibit structural defects such as cracks, perforations, corrosion, joint detachment, and pipe wall deformation, as well as functional impairments such as siltation and scaling. Failure to detect and repair these defects and impairments in a timely manner can lead to serious consequences such as pipe leaks, road collapses, environmental pollution, and even urban flooding.

[0003] Currently, pipeline inspection mainly relies on CCTV systems to acquire pipe wall images, which are then manually interpreted. Repair methods employ trenchless processes such as CIPP in-situ curing and spraying. However, existing technologies disconnect the inspection and repair processes. Inspection data requires manual analysis before a repair plan can be developed, resulting in low efficiency, reliance on experience, and a lack of intelligent decision-making capabilities. Furthermore, defect location accuracy is insufficient; relying on odometers to estimate axial position leads to significant errors, and visual sensors are prone to failure in turbid water or low-light environments. External vibration interference further exacerbates location errors, resulting in poor repair outcomes. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of the prior art, the purpose of this invention is to provide a trenchless repair method and apparatus for urban underground drainage pipes, thereby solving the problems mentioned in the background art.

[0005] The technical solution adopted by this invention to solve its technical problem is: a method for trenchless repair of urban underground drainage pipes, comprising:

[0006] The robot acquires 3D point cloud data of structured light inside the pipeline, LiDAR cross-sectional data, and distributed fiber optic acoustic sensor data, and transmits them to the digital twin server.

[0007] The digital twin server performs point cloud filtering and noise reduction on the collected data, and directly identifies the defect information and determines the three-dimensional spatial location of the defect through the geometric features of the defect region in the point cloud. The defect information includes the defect type and the coordinates of the defect in the point cloud coordinate system.

[0008] The digital twin server uses the defect information and pipeline environment parameters to call the knowledge graph and repair effect prediction model for optimization and generate repair process parameters.

[0009] The robot performs repair work according to the repair process parameters and provides real-time feedback on the quality data of the repaired layer;

[0010] The digital twin server compares the quality data with the expected indicators in the digital twin model. When the deviation exceeds a preset threshold, it adjusts the repair parameters and controls the robot to continue working.

[0011] As a further improvement of the present invention: the step of acquiring the three-dimensional point cloud data of the structured light inside the pipe, the lidar cross-sectional data, and the distributed fiber optic acoustic sensing data collected by the robot includes:

[0012] An adaptive intelligent repair robot is deployed into the pipeline. The robot is equipped with a structured light 3D imaging unit, a lidar, and a distributed fiber optic acoustic sensing unit.

[0013] As the robot moves along the pipeline, the structured light 3D imaging unit acquires 3D point cloud data of the inner wall of the pipeline through a laser emitter and an industrial camera, the lidar acquires point cloud data of the pipeline cross-section, and the distributed fiber optic acoustic sensing unit continuously captures vibration response data along the pipeline.

[0014] As a further improvement of the present invention: the digital twin server performs point cloud filtering and noise reduction processing on the collected data, and the step of directly identifying defect information and determining the three-dimensional spatial location of the defect through the geometric features of the defect region in the point cloud includes:

[0015] The digital twin server uses a statistical filtering algorithm to remove outlier noise points from the point cloud, obtaining processed 3D point cloud data. Geometric features are extracted from the processed point cloud data to identify depth abrupt change regions, curvature abnormal regions, or continuity break regions in the point cloud as candidate defect regions. The candidate defect regions are classified and labeled to obtain the defect type and the 3D spatial coordinates of the defect in the point cloud coordinate system.

[0016] As a further improvement of the present invention: the step of the digital twin server processing the collected data and determining the three-dimensional spatial location of the defect further includes:

[0017] The robot's 3D pose is associated with the point cloud coordinate system: the robot's 3D position and attitude data are obtained in real time from the inertial navigation system on the robot, and a global coordinate system is established based on the positioning data of the inertial navigation system; the 3D spatial coordinates of the defect in the point cloud coordinate system are mapped to the global coordinate system through coordinate transformation to obtain the global 3D spatial position information of the defect.

[0018] As a further improvement of the present invention: the association of the robot's three-dimensional pose with the point cloud coordinate system includes:

[0019] The point cloud density and feature point distribution uniformity of the lidar point cloud data are obtained, and the localization contribution of the point cloud data is evaluated based on the above indicators.

[0020] The gyroscope drift rate, accelerometer noise level, and odometry cumulative error of the inertial navigation system are obtained, and the instantaneous positioning accuracy of the inertial navigation system is evaluated based on the above indicators.

[0021] An energy-sensing adaptive fusion method is adopted to dynamically allocate the fusion weights of point cloud data and inertial navigation data according to the complexity of the pipeline environment and the robot's motion state;

[0022] The relative pose obtained by point cloud registration and the absolute pose calculated by inertial navigation are fused based on the fusion weight to obtain the robot's three-dimensional pose in the global coordinate system.

[0023] The point cloud observation value and inertial navigation output value at the next moment are predicted based on the 3D pose and compared with the actual collected data to obtain the fusion residual. When the fusion residual exceeds the preset threshold, the fusion weight of the corresponding data source is reduced.

[0024] As a further improvement of the present invention: the step of acquiring the robot's three-dimensional position and attitude data output in real time by the robot's inertial navigation system includes:

[0025] Using miniaturized gyroscopes and accelerometers as inertial measurement units, the robot's three-dimensional coordinates are measured directly inside the tube.

[0026] When the robot travels inside the pipeline to the location of the inspection well with known coordinates on the ground, the cumulative error of the inertial navigation system is corrected at zero speed or calibrated using ground differential GPS coordinates.

[0027] An error compensation algorithm is used to compensate for the drift error of the inertial navigation system in real time.

[0028] As a further improvement of the present invention: the step of generating repair process parameters by the digital twin server based on defect information and pipeline environment parameters, calling a knowledge graph and a repair effect prediction model for optimization, includes:

[0029] The digital twin server calls a pre-built knowledge graph of repair solutions, taking the defect type, the three-dimensional spatial location of the defect, the pipe material parameters, and the construction site environmental data as input, and calculates candidate repair solutions through a graph neural network.

[0030] For candidate repair schemes, a repair effect prediction model based on neural networks is invoked for multi-objective optimization. The objective functions of the multi-objective optimization include maximizing the bonding strength of the repair layer, minimizing the thickness non-uniformity coefficient of the repair layer, and minimizing the repair energy consumption and material cost. A non-dominated sorting genetic algorithm with an elitist strategy is used for iterative solution to obtain the Pareto optimal set of repair process parameters.

[0031] As a further improvement of the present invention: the step of the digital twin server comparing quality data with expected indicators in the digital twin model includes:

[0032] An ultrasonic phased array detection module mounted on a robot is used to collect real-time data on the interfacial bonding quality between the repair layer and the original tube wall and upload it to a digital twin server.

[0033] The interface quality data acquired in real time by the ultrasonic phased array detection module includes the repair layer thickness value measured by the ultrasonic pulse echo method and the interface bonding strength characterization value measured by the ultrasonic transmission method.

[0034] After mapping real-time quality data to the corresponding spatial location of the digital twin model, the digital twin server generates a quality deviation distribution map. When the thickness deviation or bonding strength deviation exceeds the preset threshold, the location coordinates and range of the unqualified area are automatically marked on the digital twin model.

[0035] As a further improvement of the present invention: the step of adjusting the repair parameters and controlling the robot to continue working when the deviation exceeds a preset threshold includes:

[0036] Obtain the spatial distribution and deviation amplitude of non-conforming areas with deviations exceeding a preset threshold in the digital twin model;

[0037] Based on the spatial distribution of non-conforming areas, determine the locational relationship between the currently unrepaired pipe section and the non-conforming areas;

[0038] If the non-conforming area is located at the end of the repaired pipe section and adjacent to the unrepaired pipe section, the current repair parameters are used as the initial values, and the deviation amplitude of the non-conforming area is used to calibrate the repair effect prediction model online. The calibrated model is then used to re-optimize the repair process parameters of the remaining unrepaired pipe section, generate correction parameters, and issue them to the robot.

[0039] If a non-conforming area is isolated within a repaired pipe section and is not adjacent to an unrepaired pipe section, the non-conforming area will be marked as a repair point. After the entire pipe section is repaired, the robot will return to the repair point to perform local secondary repair.

[0040] On the other hand, the present invention also provides a trenchless repair device for urban underground drainage pipes to implement the method described in any of the preceding claims, comprising:

[0041] An adaptive intelligent repair robot, and a digital twin server that communicates with the adaptive intelligent repair robot;

[0042] The repair robot includes:

[0043] The adaptive walking mechanism can adjust its own contour according to changes in the inner diameter of the pipe to maintain stable movement;

[0044] The multi-source sensing unit integrates a structured light 3D imaging unit, a lidar, an inertial navigation system, and a distributed fiber optic acoustic sensing unit, used to collect 3D point cloud data inside the pipeline, cross-sectional point cloud data, robot pose data, and vibration response data along the pipeline.

[0045] The work execution unit connects to the robot body through a standardized interface and is used to perform at least one of the following tasks: pipeline cleaning, obstacle breaking, repair material application, or repair quality inspection.

[0046] The control unit is used to process the sensed data in real time and adjust the working parameters of the work execution unit.

[0047] The digital twin server includes: a digital twin modeling module, used to construct a three-dimensional digital model of the pipeline structure based on the perception data returned by the robot;

[0048] The repair module has a repair knowledge base and a repair effect prediction model, which is used to generate repair plans and process parameters based on defect information and environmental parameters.

[0049] The correction module is used to perform multi-objective optimization of the repair process parameters and adjust the process parameters based on real-time feedback quality data during the repair process.

[0050] Compared with the prior art, the beneficial effects of the present invention are:

[0051] This invention employs structured light 3D imaging and lidar to directly acquire point cloud data of the pipeline's inner wall. Defects are directly identified through the geometric features of the point cloud, and their 3D coordinates in the point cloud coordinate system are provided. This is then combined with global coordinate mapping from an inertial navigation system, avoiding indirect conversion errors from 2D images to 3D space. Simultaneously, a distributed fiber optic acoustic sensing unit can capture external vibration characteristics and compensate for and adjust the inertial navigation data. An energy-sensing adaptive fusion method dynamically allocates fusion weights between the point cloud and inertial navigation data based on the complexity of the pipeline environment and the robot's motion state, improving the accuracy of defect localization. The defect identification results are directly input into the repair scheme knowledge graph, and candidate repair schemes are calculated through graph neural networks. Then, the repair effect prediction model constructed by the neural network is used for multi-objective optimization to generate Pareto optimal repair process parameters, eliminating the reliance on human experience. During the repair process, the ultrasonic phased array detection module collects real-time data on the repair layer thickness and interface bonding strength. The digital twin server compares the measured quality with the model prediction indicators point by point. When the deviation exceeds the threshold, Bayesian linear regression is used to calibrate the repair effect prediction model online based on the location correlation of the non-conforming areas, and optimize the process parameters. This effectively solves the technical problems of low defect identification rate, poor positioning accuracy, and unstable repair quality in traditional technologies. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating a trenchless repair method for urban underground drainage pipes provided in this application. Detailed Implementation

[0053] To enable a clear and complete understanding of the technical solution, the present invention will now be further described in conjunction with the embodiments and accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0054] It should be understood that, when used in this specification and the appended claims, the terms "comprising" and "including" indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0055] It should also be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0056] It should also be further understood that the term "and / or" as used in this specification and the appended claims refers to any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0057] In traditional trenchless repair of urban underground drainage pipelines, the detection and repair processes are disconnected. Repair plan formulation relies on manual experience analysis of detection data, lacking data-driven intelligent decision-making capabilities. At the same time, the defect location accuracy is insufficient, mainly relying on odometers to estimate axial position, resulting in large positioning errors. Furthermore, visual sensors are prone to failure in turbid water or dark environments, and external vibration interference further exacerbates the positioning error, affecting the accuracy of defect identification and the reliability of repair operations, thereby reducing the overall efficiency and quality of pipeline repair.

[0058] During the inspection of old urban drainage pipelines, the CCTV system caused images to become blurred due to silt deposits inside the pipes, making it impossible to clearly capture details of cracks in the pipe walls. Operators then relied on odometers to estimate the axial position of the defects. However, due to external construction vibrations in the pipe bends, the actual defect coordinates deviated from the estimated position, causing the subsequent repair robot to fail to accurately cover the target area and the repair material to fail to effectively fill the cracks, resulting in local repair failure.

[0059] If the above problems are not addressed, structural defects in pipelines may not be identified and repaired in a timely and accurate manner, leading to a chain of technical consequences such as leakage, road collapse, environmental pollution, and even urban flooding. This poses a continuous threat to the long-term operational safety of urban infrastructure, while also increasing subsequent maintenance costs and shortening the service life of pipelines.

[0060] Embodiments of the present invention provide a trenchless repair method for urban underground drainage pipes, including...

[0061] The robot acquires 3D point cloud data of structured light inside the pipeline, LiDAR cross-sectional data, and distributed fiber optic acoustic sensor data, and transmits them to the digital twin server.

[0062] The digital twin server performs point cloud filtering and noise reduction on the collected data, and directly identifies the defect information and determines the three-dimensional spatial location of the defect through the geometric features of the defect region in the point cloud. The defect information includes the defect type and the coordinates of the defect in the point cloud coordinate system.

[0063] The digital twin server uses the defect information and pipeline environment parameters to call the knowledge graph and repair effect prediction model for optimization and generate repair process parameters.

[0064] The robot performs repair work according to the repair process parameters and provides real-time feedback on the quality data of the repaired layer;

[0065] The digital twin server compares the quality data with the expected indicators in the digital twin model. When the deviation exceeds a preset threshold, it adjusts the repair parameters and controls the robot to continue working.

[0066] For ease of understanding, the following explains some key terms in this embodiment:

[0067] Trenchless repair of urban underground drainage pipes refers to a technology that uses robots and other equipment to enter the interior of underground drainage pipes without large-scale road excavation to detect, assess, and repair structural defects or functional malfunctions. This method aims to reduce the impact on urban traffic and the environment while improving repair efficiency.

[0068] In this embodiment, the robot specifically refers to an adaptive intelligent repair robot capable of autonomously navigating, sensing, operating, and providing data feedback in complex pipeline environments. This robot integrates multiple sensors and actuators, serving as the core equipment for achieving intelligent repair.

[0069] Structured light 3D point cloud data is a high-precision point cloud data of the pipe's inner wall surface obtained through structured light 3D imaging technology using lidar. It provides the geometric contour of the pipe cross-section and is crucial for identifying defects such as pipe deformation and misalignment. The data is represented as a set of discrete points, each containing its coordinates (X, Y, Z) in 3D space and its possible corresponding reflection intensity, accurately reflecting the morphology and defect characteristics of the pipe's inner wall.

[0070] Distributed fiber optic acoustic sensing data consists of vibration response signals continuously captured along the length of a pipeline using distributed fiber optic acoustic sensing technology. This data reflects acoustic and vibration events in the environment inside and outside the pipeline and can be used to identify external disturbances or anomalies inside the pipeline.

[0071] A digital twin server is a computing platform that integrates data processing, model building, intelligent decision-making, and control command issuance. Its core function is to build and maintain a digital twin model of the pipeline, enabling real-time mapping, analysis, and prediction of the physical pipeline.

[0072] Point cloud filtering and noise reduction refers to the preprocessing of raw point cloud data to remove outliers, noise points, or redundant data, thereby improving the quality of the point cloud data and the accuracy of subsequent processing. Common processing methods include statistical filtering and radius filtering.

[0073] Defect information refers to a detailed description of pipeline defects identified through analysis of internal pipeline data. This information typically includes the type of defect (such as cracks, perforations, corrosion, etc.) and its precise location in a specific coordinate system.

[0074] The three-dimensional spatial location of a defect refers to its precise three-dimensional coordinates within the pipe or in the global coordinate system. Accurate defect location information is fundamental for developing repair plans and guiding robot operations.

[0075] Pipeline environmental parameters refer to environmental factors and pipeline properties related to pipeline repair work. These include, but are not limited to, pipeline material, diameter, burial depth, water flow conditions, temperature, humidity, and surrounding soil conditions.

[0076] A knowledge graph is a structured knowledge base that represents knowledge graphically. In this embodiment, the repair scheme knowledge graph stores a large amount of information on pipeline defects, repair technologies, repair materials, equipment models, and the relationships and applicability rules between them, which can support intelligent reasoning and decision-making.

[0077] A repair outcome prediction model is a predictive tool based on data-driven or physical model construction. This model can predict key quality indicators of the repaired pipeline, such as bond strength, thickness uniformity, and service life, based on input repair process parameters and pipeline environmental conditions.

[0078] Repair process parameters refer to the various operational parameters that need to be set during pipeline repair operations. These include robot travel speed, repair material flow rate, curing temperature, curing pressure, and curing time, and these parameters directly affect the repair quality and efficiency.

[0079] Repair work refers to the actual repair operations performed by a robot inside a pipeline based on generated repair process parameters. This may include pipeline cleaning, defect pretreatment, spraying of repair materials, or curing of linings.

[0080] Repair layer quality data refers to the various physical and mechanical performance indicators of the repair layer collected in real time by sensors during or after the repair operation. Examples include the thickness of the repair layer and its adhesion strength to the original pipe wall.

[0081] A digital twin model is a virtual mapping of a physical pipeline in digital space. This model not only includes the pipeline's geometry but also integrates its material properties, defect information, environmental parameters, and real-time operating and repair status, enabling simulation, analysis, and prediction.

[0082] A preset threshold is a pre-set judgment standard value during system operation. When the deviation of a certain monitoring indicator exceeds the threshold, the system will trigger a corresponding alarm or control action.

[0083] This embodiment provides a trenchless repair method for urban underground drainage pipelines, which achieves intelligent repair operations by integrating multi-source sensing, digital twin modeling, intelligent decision-making, and real-time feedback control.

[0084] This method involves acquiring 3D point cloud data of the pipe's interior using structured light, LiDAR cross-sectional data, and distributed fiber optic acoustic sensing data collected by a robot, and transmitting this data to a digital twin server. In one implementation, the robot can be equipped with independent structured light sensors, LiDAR sensors, and distributed fiber optic acoustic sensors, each collecting its respective type of data. For example, the structured light sensor can employ a monocular or binocular vision system to acquire 3D point clouds by projecting specific patterns and capturing their deformation; the LiDAR sensor can acquire pipe cross-sectional data using rotational or line scanning; and the distributed fiber optic acoustic sensor can sense vibrations through optical fibers laid on the pipe's inner wall or integrated into the robot's cable. This data can be transmitted to the digital twin server in real time via wired or wireless communication modules. In another implementation, the robot can be equipped with an integrated multispectral imaging system that simultaneously acquires data similar to structured light and LiDAR in a single scan using light sources and detectors of different wavelengths, and acquires vibration data through additional acoustic sensors.

[0085] The digital twin server performs point cloud filtering and noise reduction on the collected data. It directly identifies defect information and determines the three-dimensional spatial location of defects by analyzing the geometric features of defect regions within the point cloud. Defect information includes the defect type and its coordinates in the point cloud coordinate system. Specifically, the server can use statistical analysis methods to filter the point cloud data. For example, it can calculate the average distance to neighboring points of each point and remove outliers with excessively large distances. When identifying defects, the server can analyze the local geometric characteristics of the point cloud data, such as abrupt changes in the surface normal vector, abnormal changes in curvature, or discontinuities in point cloud density. For instance, when a crack appears in the inner wall of a pipe, it will appear as a linear region with a sharp change in depth or curvature in the point cloud data; when a perforation occurs, it will appear as a localized missing or void in the point cloud data. The server uses preset geometric feature templates or algorithms to identify these abnormal regions as candidate defect regions and further classifies them into specific defect types, such as cracks, corrosion pits, or interface detachment. At the same time, the server can calculate the center point coordinates or bounding box coordinates of these defect areas, thereby determining the three-dimensional spatial position of the defects in the point cloud coordinate system.

[0086] The digital twin server, based on defect information and pipeline environmental parameters, utilizes a knowledge graph and a repair effect prediction model for optimization, generating repair process parameters. In one implementation, the server can pre-build a database containing various pipeline defects, repair technologies, repair materials, and their applicability rules. Upon receiving defect information and pipeline environmental parameters, the server can match them according to preset rules and retrieve several candidate repair solutions from the database. For example, for circumferential cracks in concrete pipelines, candidate solutions might include CIPP in-situ curing or spray repair. For each candidate solution, the server can invoke a repair effect prediction model built based on empirical formulas or statistical regression models, inputting different combinations of process parameters to predict post-repair bond strength, thickness uniformity, and other indicators. By comparing these prediction results, the server can select the optimal combination of process parameters under specific constraints.

[0087] The robot performs repair work according to the repair process parameters and provides real-time feedback on the quality data of the repair layer. In one implementation, the robot can be equipped with different work execution units, such as a spraying module or a CIPP curing module. Upon receiving the repair process parameters from the server, the robot's control unit drives the work execution units, such as adjusting the nozzle angle, spraying pressure, and material flow rate of the spraying module, and controlling the robot's walking speed to ensure that the repair material is uniformly coated on the defective area. During the repair process, the robot can be equipped with simple vision sensors or contact probes to monitor the initial quality of the repair layer in real time. For example, it can determine the uniformity of the spraying through image analysis or determine the initial thickness of the repair layer through contact measurement, and transmit this data back to the digital twin server in real time.

[0088] The digital twin server compares quality data with expected indicators in the digital twin model. When the deviation exceeds a preset threshold, it adjusts repair parameters and controls the robot to continue operation. Specifically, the digital twin server maintains a digital twin model of the pipeline, which includes the pipeline's state before repair, the expected geometry of the repair layer, and material performance indicators. When it receives real-time repair layer quality data from the robot, the server maps this data to the corresponding spatial location in the digital twin model. The server can calculate the deviation between the actual quality data and the expected indicators in the digital twin model. For example, if the actual repair layer thickness in a certain area is lower than the expected thickness, or the initial bonding strength does not reach the expected value, and the deviation exceeds a preset allowable threshold, the server will determine that area as a non-conforming area. At this time, the server can automatically adjust the process parameters of subsequent repair operations based on the magnitude and type of the deviation. For example, it can increase the flow rate of the sprayed material or reduce the robot's walking speed, and issue new control commands to the robot to perform secondary repairs on the non-conforming area or correct the parameters of subsequent pipe sections to ensure that the overall repair quality meets the requirements.

[0089] In one embodiment of the present invention, the step of acquiring the three-dimensional point cloud data of the structured light inside the pipe, the lidar cross-sectional data, and the distributed fiber optic acoustic sensor data collected by the robot includes:

[0090] An adaptive intelligent repair robot is deployed into the pipeline. The robot is equipped with a structured light 3D imaging unit, a lidar, and a distributed fiber optic acoustic sensing unit.

[0091] As the robot moves along the pipeline, the structured light 3D imaging unit acquires 3D point cloud data of the inner wall of the pipeline through a laser emitter and an industrial camera, the lidar acquires point cloud data of the pipeline cross-section, and the distributed fiber optic acoustic sensing unit continuously captures vibration response data along the pipeline.

[0092] As the robot moves along the pipeline, the structured light 3D imaging unit, working in conjunction with a laser emitter and an industrial camera, acquires high-precision 3D point cloud data of the pipeline's inner wall. This provides a foundation for identifying the geometric features of internal pipeline defects. Simultaneously, the lidar independently acquires fine point cloud data of the pipeline cross-section, supplementing the accuracy of structured light in cross-sectional dimension measurement. This is crucial for identifying defects such as pipeline deformation and interface misalignment. Furthermore, the distributed fiber optic acoustic sensing unit continuously captures vibration response data along the pipeline to monitor external environmental disturbances, providing additional risk assessment data for repair operations. This allows the robot to simultaneously acquire pipeline information from different dimensions in a single entry, avoiding the cumbersome process and spatial positioning errors associated with traditional methods that require multiple entries and deployments of different detection equipment.

[0093] Furthermore, the specific methods for the structured light 3D imaging unit to acquire 3D point cloud data of the pipe's inner wall include:

[0094] The structured light 3D imaging unit projects an coded grating pattern onto the inner wall surface of the pipe. The coded grating pattern is generated using a three-step phase-shifting method.

[0095] An industrial camera captures deformable grating images modulated by the inner wall of a pipe, and the absolute phase value of each pixel is calculated using a phase unwrapping algorithm.

[0096] Based on the principle of triangulation, the absolute phase value is converted into the three-dimensional spatial coordinates corresponding to the pixel, and three-dimensional point cloud data of the inner wall of the pipe is generated point by point. Each point cloud data contains the X-axis coordinate, Y-axis coordinate, Z-axis coordinate and the corresponding reflection intensity value of the point.

[0097] This embodiment utilizes a structured light 3D imaging unit to project coded grating patterns onto the inner wall surface of a pipe. These coded grating patterns are generated using a three-step phase-shifting method to ensure the accuracy of the coded information within the patterns. When these patterns are projected onto the inner wall of the pipe, the grating patterns deform due to the geometric undulations of the inner wall. An industrial camera then acquires these deformed grating images modulated by the inner wall of the pipe; these images record the modulation information of the grating by the inner wall contour. The acquired deformed grating images are processed using a phase unwrapping algorithm to calculate the absolute phase value of each pixel. The phase unwrapping algorithm eliminates phase ambiguity, ensuring that each pixel corresponds to a unique phase value, thus accurately reflecting the undulations of the pipe wall. Based on the principle of triangulation, these absolute phase values ​​are converted into three-dimensional spatial coordinates corresponding to each pixel, and three-dimensional point cloud data of the inner wall of the pipe is generated point by point. Each point cloud data not only contains the X-axis, Y-axis, and Z-axis coordinates of that point but also additionally contains the corresponding reflection intensity value. The triangulation principle, combined with absolute phase values, can accurately convert two-dimensional image information into three-dimensional spatial information, while the point-by-point generation method ensures that the point cloud can completely cover the entire inner wall of the pipe.

[0098] In addition, specific methods for the distributed fiber optic acoustic sensing unit to continuously capture vibration response data along the pipeline include:

[0099] Distributed optical fiber acoustic sensing fiber is laid axially along the inner wall of the pipe or integrated into the tail cable of the robot, and the fiber is kept in coupling contact with the inner wall of the pipe.

[0100] The distributed fiber acoustic sensing unit uses a pulsed laser to emit coherent light pulses into the optical fiber and senses vibration signals along the fiber by detecting the phase change of the backscattered Rayleigh light in the fiber.

[0101] Vibration data is continuously collected along the length of the optical fiber at a preset spatial sampling interval. Each sampling point records the vibration waveform in the time domain, forming a three-dimensional vibration response dataset containing the time axis, spatial axis, and vibration amplitude.

[0102] Distributed fiber optic acoustic sensing units are the core devices for sensing vibration signals along a fiber optic line. Their function is to transmit and receive optical signals, and process the received signals to extract vibration information. They typically consist of a pulsed laser, an optical coupler, a photodetector, a data acquisition module, and a signal processing unit. The pulsed laser generates short-duration, high-energy coherent optical pulses. These pulses, propagating in the optical fiber, excite Rayleigh scattering along the line. Coherent optical pulses are characterized by a fixed phase relationship and the same frequency, maintaining good interference characteristics during propagation in the fiber, thus ensuring stable phase information in the backscattered Rayleigh light. When the optical fiber is subjected to external vibration, its local strain and refractive index undergo slight changes, causing a phase change in the backscattered Rayleigh light. By detecting these phase changes with high precision, the distributed fiber optic acoustic sensing unit can infer the vibration signal along the fiber optic line.

[0103] The sensing fiber arrangement provided in this embodiment involves laying the fiber optic cable axially along the inner wall of the pipe or integrating it into the robot's tail cable, ensuring that the fiber maintains coupling contact with the inner wall of the pipe. This ensures that vibrations around the pipe can be effectively transmitted to the fiber, avoiding signal distortion. A pulsed laser emits coherent light pulses, and vibrations along the pipeline are sensed by detecting the phase change of the backscattered Rayleigh light. Utilizing the sensitivity of Rayleigh scattering light to vibration changes, minute vibrations along the fiber optic cable can be captured, enabling continuous vibration sensing along long-distance pipelines. Vibration data is continuously collected at preset spatial sampling intervals, and the time-domain vibration waveform of each sampling point is recorded to form a three-dimensional vibration response dataset. This ensures the regularity of the data, allowing the spatial location, time, and intensity of vibration events to be recorded.

[0104] In one embodiment of the present invention, the digital twin server performs point cloud filtering and noise reduction processing on the collected data, and the step of directly identifying defect information and determining the three-dimensional spatial location of the defect through the geometric features of the defect region in the point cloud includes:

[0105] The digital twin server uses a statistical filtering algorithm to remove outlier noise points from the point cloud, obtaining processed 3D point cloud data;

[0106] After receiving the 3D point cloud data of the pipe interior structured light, the LiDAR cross-sectional data, and the distributed fiber acoustic sensing data collected by the robot, the digital twin server uses a statistical filtering algorithm to preprocess the point cloud data to effectively filter out these noise points that do not represent the true geometric features of the pipe, thereby obtaining purer and more accurate 3D point cloud data.

[0107] Geometric features are extracted from the processed point cloud data to identify depth abrupt change regions, curvature abnormal regions, or continuity break regions in the point cloud as candidate defect regions.

[0108] The digital twin server performs in-depth geometric feature extraction on the processed point cloud data. Since pipe defects inevitably exhibit geometric anomalies—for example, cracks cause abrupt changes in depth and curvature anomalies, perforations cause discontinuity breaks, and corrosion pits may simultaneously exhibit abrupt changes in depth and curvature anomalies—identifying regions of abrupt changes in depth, anomalies in curvature, or discontinuity breaks in the point cloud can effectively distinguish potential defect areas from normal pipe walls, forming defect candidate regions. Specifically:

[0109] The normal vector and curvature value of each point in the point cloud are calculated. The curvature value is obtained by fitting a local quadratic surface. The normal vector represents the orientation of the point cloud surface in a local region and can be determined by analyzing the three-dimensional point distribution of each point in its neighborhood. The curvature value quantifies the degree of curvature of the point cloud surface and is a geometric feature that distinguishes normal pipe walls from defective areas.

[0110] Points with curvature values ​​exceeding a first preset threshold are marked as curvature anomaly points, and areas formed by consecutive curvature anomaly points are marked as curvature anomaly regions. The first preset threshold is an empirical or statistical critical value used to distinguish between normal pipe undulations and bends caused by defects. When the curvature value of a point exceeds this threshold, it indicates that there is abnormal bending or deformation in the local area where that point is located.

[0111] The gradient change of adjacent points in the point cloud in the depth direction is calculated, and the region where the depth gradient exceeds the second preset threshold is marked as a depth change region. The depth direction usually refers to the direction perpendicular to the inner wall surface of the pipe. The change in depth gradient measures the severity of the point cloud in the depth dimension, which can effectively capture defects with obvious depth differences such as cracks and perforations.

[0112] Connectivity analysis is performed on point clouds to identify spatial discontinuities or gaps. Regions that do not meet preset connectivity conditions are marked as discontinuous regions. Connectivity analysis aims to detect unexpected blank areas in the point cloud data, which may correspond to defects such as missing pipe walls, breaks, or disconnected interfaces. Preset connectivity conditions may include minimum point count, minimum area, or minimum volume; regions that do not meet these conditions are considered discontinuous regions.

[0113] The union of curvature anomaly regions, depth abrupt change regions, and continuity break regions is used as the defect candidate region;

[0114] The candidate defect regions are classified and labeled to obtain the defect type and its three-dimensional spatial coordinates in the point cloud coordinate system. Specifically:

[0115] Extract the geometric feature vector of the defect candidate region. The geometric feature vector includes the region area, region shape factor, region depth, region aspect ratio, region principal axis direction, region normal vector uniformity, and region point cloud density.

[0116] The extracted geometric feature vectors are input into a pre-trained classifier or deep learning model, and the classifier or deep learning model outputs a defect type label and a corresponding confidence level. The defect type includes at least one of cracks, perforations, corrosion pits, interface detachment, and pipe wall deformation.

[0117] Calculate the coordinates of the center point of the defect candidate region as the three-dimensional spatial coordinates of the defect in the point cloud coordinate system, or calculate the minimum bounding rectangle of the defect candidate region and record its vertex coordinates as the spatial range of the defect.

[0118] This embodiment can classify and locate identified defect candidate regions in three-dimensional space. By extracting multi-dimensional geometric feature vectors and inputting them into a pre-trained classifier or deep learning model, it achieves automated and intelligent identification of defect types, improving identification efficiency and accuracy and reducing reliance on human experience. Simultaneously, it provides a method for determining the three-dimensional spatial coordinates of defects. Whether obtaining the coordinates of the center point or the vertex coordinates of the minimum bounding rectangle, it can provide spatial positioning information for subsequent repair operations. This allows the digital twin server to obtain high-quality defect baseline data, enabling it to more accurately call upon knowledge graphs and repair effect prediction models for optimization based on this detailed defect information and pipeline environment parameters. This generates highly targeted and effective repair process parameters, thereby guiding the robot to perform precise repair operations.

[0119] In one embodiment of the present invention, the step of the digital twin server processing the collected data and determining the three-dimensional spatial location of the defect further includes:

[0120] Associating the robot's 3D pose with the point cloud coordinate system: obtaining the robot's 3D position and attitude data output in real time by the robot's inertial navigation system, and establishing a global coordinate system based on the positioning data of the inertial navigation system;

[0121] The three-dimensional spatial coordinates of the defect in the point cloud coordinate system are mapped to the global coordinate system through coordinate transformation to obtain the global three-dimensional spatial location information of the defect.

[0122] Associating the robot's 3D pose with the point cloud coordinate system involves establishing a transformation relationship between the robot's own coordinate system and the local coordinate system of the point cloud data acquired by the structured light 3D imaging unit. This can be achieved through pre-calibration. During robot manufacturing or assembly, the installation position and attitude of the structured light 3D imaging unit relative to the inertial navigation system reference point are measured, thus obtaining a fixed rigid body transformation matrix. Alternatively, it can be achieved through online registration algorithms. For example, during robot movement, feature points in the point cloud data are matched with the robot's kinematic model to estimate the relative pose between them in real time.

[0123] This embodiment associates the robot's 3D pose with a point cloud coordinate system and establishes a global coordinate system based on the positioning data of the inertial navigation system. This transforms the 3D spatial coordinates of the defect in the point cloud coordinate system to the global coordinate system, allowing the defect location information, originally existing only in a local point cloud coordinate system, to be accurately converted to a unified global coordinate system. The robot acquires structured light 3D point cloud data, lidar cross-sectional data, and distributed fiber optic acoustic sensor data from inside the pipe, and transmits this data to a digital twin server. The structured light 3D imaging unit on the robot acquires 3D point cloud data of the pipe's inner wall, and the digital twin server processes the acquired data. By directly identifying the defect information and determining its 3D spatial location through the geometric features of the defect region in the point cloud, this embodiment converts these local defect location information into global location information, solving the problem of the lack of a unified global reference for defect locations in local coordinate systems. In this way, the digital twin server can obtain the global 3D spatial location information of the defect throughout the entire pipe.

[0124] In one embodiment of the present invention, associating the robot's three-dimensional pose with the point cloud coordinate system includes:

[0125] The point cloud density and feature point distribution uniformity of the lidar point cloud data are obtained, and the localization contribution of the point cloud data is evaluated based on the point cloud density and feature point distribution uniformity of the lidar point cloud data.

[0126] The gyroscope drift rate, accelerometer noise level, and odometry cumulative error of the inertial navigation system are obtained, and the instantaneous positioning accuracy of the inertial navigation system is evaluated based on these parameters.

[0127] An energy-sensing adaptive fusion method is adopted to dynamically allocate the fusion weights of point cloud data and inertial navigation data according to the complexity of the pipeline environment and the robot's motion state;

[0128] The relative pose obtained by point cloud registration and the absolute pose obtained by inertial navigation calculation are fused based on the fusion weight to obtain the robot's three-dimensional pose in the global coordinate system.

[0129] The point cloud observation value and inertial navigation output value at the next moment are predicted based on the 3D pose and compared with the actual collected data to obtain the fusion residual. When the fusion residual exceeds the preset threshold, the fusion weight of the corresponding data source is reduced.

[0130] As the robot moves along the pipeline, the system continuously acquires the point cloud density and feature point distribution uniformity of the LiDAR point cloud data, which directly reflects the quality of the point cloud data itself and the reliability of its positioning. Simultaneously, the system also acquires the gyroscope drift rate, accelerometer noise level, and odometry cumulative error of the inertial navigation system in real time, reflecting the current error level and instantaneous positioning accuracy of the inertial navigation system.

[0131] Based on this, the system adopts an energy-sensing adaptive fusion method to calculate and allocate fusion weights for point cloud data and inertial navigation data according to the complexity of the current pipeline environment and the robot's motion state.

[0132] Subsequently, based on these dynamically assigned fusion weights, the system fuses the robot's relative pose information obtained through point cloud registration with the robot's absolute pose information calculated by the inertial navigation system. Point cloud registration provides local relative position information, while inertial navigation provides continuous global absolute position information. Through weighted fusion, the robot's three-dimensional pose in the global coordinate system is obtained.

[0133] The system predicts the point cloud observations and inertial navigation outputs for the next moment based on the currently estimated 3D pose. These predicted values ​​are compared with the actually acquired data to calculate the fusion residual. If the fusion residual of a data source exceeds a preset threshold, it indicates that the actual reliability of that data source may be lower than expected, or that its data quality has deteriorated. In this case, the system will promptly reduce the fusion weight of the corresponding data source to correct unreasonable weight allocation and prevent low-quality data from negatively impacting the overall positioning results.

[0134] This embodiment effectively addresses the problem of insufficient positioning accuracy in traditional pose fusion methods when the pipeline environment becomes complex and the robot's motion state changes. By evaluating the reliability of LiDAR point cloud data and inertial navigation system data in real time, and dynamically adjusting the fusion weights according to the complexity of the pipeline environment and the robot's motion state, the system can always utilize the most reliable data source for pose estimation.

[0135] In one embodiment, the specific implementation steps of the energy-sensing adaptive fusion method include:

[0136] Define the pipeline environment complexity index Environmental complexity index The calculation formula is derived from a comprehensive analysis of the pipe bend curvature, pipe diameter change rate, and point cloud feature richness. ,in This is the normalized value of the curve curvature. This is the normalized value of the pipe diameter change rate. The normalized value of the point cloud feature density. , , Preset weighting coefficients;

[0137] Define the robot motion state index Motion status index The value is determined based on the robot's current rate of change of velocity and acceleration, ranging from 0 to 1, where it is used during uniform linear motion. Take the minimum value when accelerating, decelerating, or turning. Take the maximum value;

[0138] According to the environmental complexity index and motion status index Calculate the fusion weights of point cloud data Fusion weights with inertial navigation data The calculation formula is: , , where α and β are preset weight coefficients and satisfy α+β=1;

[0139] when When the value exceeds the first environmental threshold, reduce the fusion weight of the point cloud data to 50% of the initial weight;

[0140] When M_state is higher than the first motion threshold, the fusion weight of the inertial navigation data is increased to 120% of the initial weight.

[0141] In one embodiment of the present invention, the step of acquiring the robot's three-dimensional position and attitude data output in real time by the robot's inertial navigation system includes:

[0142] Using miniaturized gyroscopes and accelerometers as inertial measurement units, the robot's three-dimensional coordinates are measured directly inside the tube.

[0143] When the robot travels inside the pipeline to the location of the inspection well with known coordinates on the ground, the cumulative error of the inertial navigation system is corrected at zero speed or calibrated using ground differential GPS coordinates.

[0144] An error compensation algorithm is used to compensate for the drift error of the inertial navigation system in real time.

[0145] Furthermore, the specific implementation steps of the error compensation algorithm include:

[0146] An error state model for an inertial navigation system is established, which includes state variables such as position error, velocity error, attitude error, gyroscope bias error, and accelerometer bias error.

[0147] The zero-velocity correction method is adopted. When the robot is detected to be stationary, zero velocity is used as the observation value and input into the Kalman filter to estimate and correct the error state.

[0148] The heading correction method is adopted. When the robot passes through a straight pipe section with a known heading, the heading angle error of the inertial navigation system is corrected by using the theoretical heading angle as a reference value.

[0149] When the robot passes a manhole whose ground differential GPS coordinates are known, the coordinates of the manhole are used as position observations to correct the position error of the inertial navigation system.

[0150] The calculation period for error compensation is 1-10 seconds, and the error state variables are reinitialized to zero after correction.

[0151] This embodiment utilizes miniaturized gyroscopes and accelerometers as inertial measurement units, enabling the robot to perform autonomous 3D coordinate measurements in confined pipe environments. As the robot travels along the pipe, the inertial measurement units continuously output angular velocity and linear acceleration data, which are integrated and attitude calculated in real-time by the robot's internal processor to determine its current 3D position and attitude. When the robot reaches a pre-set manhole location with known ground coordinates, the system triggers error correction. At this point, the manhole coordinates obtained from the ground differential GPS system are used to perform zero-velocity correction or coordinate calibration for the long-term accumulated drift error of the inertial navigation system, effectively preventing error accumulation and ensuring the accuracy of the positioning reference. Between two manhole corrections, the system continuously runs an error compensation algorithm, such as using Kalman filtering to estimate and correct the error state of the inertial navigation system in real time, thereby suppressing drift caused by factors such as gyroscope zero bias and accelerometer zero bias. Combined with the step of associating the robot's 3D pose with the point cloud coordinate system, this solution can provide accurate global coordinates for the 3D spatial location of defects inside the pipe, providing a reliable positioning basis for generating repair process parameters and robot repair operations.

[0152] In one embodiment of the present invention, the digital twin server, based on defect information and pipeline environment parameters, calls a knowledge graph and a repair effect prediction model for optimization to generate repair process parameters, including the following steps:

[0153] The digital twin server calls a pre-built knowledge graph of repair solutions, taking the defect type, the three-dimensional spatial location of the defect, the pipe material parameters, and the construction site environmental data as input, and calculates candidate repair solutions through a graph neural network.

[0154] For each candidate repair scheme, a repair effect prediction model based on a neural network is invoked for multi-objective optimization to obtain the Pareto optimal set of repair process parameters.

[0155] The digital twin server first invokes a pre-built knowledge graph of repair solutions. This knowledge graph stores a vast amount of historical repair experience and expert knowledge, linking various defect types, pipe materials, environmental conditions, and corresponding repair technologies, materials, and equipment in a graph structure. When it receives the identified defect type, the defect's 3D spatial location, pipe material parameters, and construction site environmental data as input, the digital twin server uses a graph neural network to reason through the knowledge graph. After obtaining candidate repair solutions, it invokes a repair effect prediction model built on a neural network to virtually evaluate the process parameters corresponding to each candidate solution. Through multi-objective optimization, it can explore a Pareto optimal solution set that achieves the best balance among different objectives. From this Pareto optimal solution set, the repair process parameter set with the highest overall satisfaction is selected and used as the instruction to guide the robot in performing the repair operation.

[0156] The methods for constructing a knowledge graph of repair solutions include:

[0157] Using a graph database as the storage medium, a set of nodes and a set of edges are established. The set of nodes includes: disease type nodes, pipe material nodes, pipe diameter nodes, burial depth nodes, repair technology nodes, repair material nodes, and equipment model nodes.

[0158] The edge set stores the association relationships and applicability weights between nodes. The edge weight between the disease type node and the repair technology node represents the applicability of the repair technology to the disease type, and the value range is 0-1.

[0159] The initial weights are determined using the analytic hierarchy process (AHP), a judgment matrix A is constructed, and the largest eigenvalue is calculated. The corresponding normalized feature vector is used as the initial weight vector;

[0160] Collect historical restoration project case data. Each case includes an input feature vector and a label of the actual restoration scheme used. Use a logistic regression model to iteratively correct the edge weights of the knowledge graph.

[0161] When performing graph neural network calculations, the actual disease feature vector and the on-site environmental parameter vector are used as queries. A graph attention network is used to calculate the matching degree between the query and each candidate repair scheme node. The matching degree calculation formula is as follows: ,in Here, W represents the attention coefficient, and W is the weight matrix. and These are the feature representations of adjacent nodes and the current node, respectively. This is the activation function.

[0162] This embodiment constructs a structured knowledge graph of repair solutions and determines initial weights using the analytic hierarchy process (AHP). Then, iteratively corrects the edge weights using a logistic regression model based on historical engineering case data, enabling the knowledge graph weights to more accurately reflect actual engineering experience. Furthermore, a graph attention network is employed during inference, which adaptively allocates attention based on actual defect characteristics and on-site environmental parameters. This allows for more precise calculation of the matching degree between the query and each candidate repair solution node, effectively addressing the problems of inaccurate weights and non-adaptive matching in traditional knowledge graphs, and improving the adaptability of candidate repair solutions obtained through knowledge graph inference.

[0163] In one implementation, the training method for the repair effect prediction model includes:

[0164] The repair effect prediction model adopts a physical information neural network architecture. The input layer receives the process parameter vector X = [v, q, T, p, t], where v is the robot walking speed, q is the repair material flow rate, T is the curing temperature, p is the curing pressure, and t is the curing time.

[0165] The hidden layers use a 5-layer fully connected network with 256 neurons per layer. Each layer is followed by a batch normalization layer and a Dropout layer. The Dropout rate is 0.2, and the activation function is the Swish function.

[0166] Output layer outputs repair effect vector ,in For the predicted value of bond strength, For thickness non-uniformity coefficient, Energy consumption per unit length For the expected service life;

[0167] The network's loss function includes a data fitting term and a constraint term:

[0168] ;

[0169] in For the residuals of the curing kinetic equation, For the residuals of the heat conduction equation, and Here, N is a hyperparameter, representing the total number of training samples.

[0170] The training data comes from finite element simulation synthetic data and historical repair project measured data. The Adam optimizer is used for training with an initial learning rate of 0.001. The learning rate decays to 0.9 times every 50 epochs, and a total of 500 epochs are trained.

[0171] The repair effect prediction model in this embodiment effectively addresses the issues of physical inconsistency and insufficient accuracy that traditional models may encounter when predicting repair effects. It applies a physical information neural network architecture to repair effect prediction by incorporating the residuals of the curing kinetics equation and the heat conduction equation into the loss function. Based on this, it provides a more accurate and reliable assessment of repair effects for the digital twin server when generating repair process parameters. During multi-objective optimization, the model can accurately predict bond strength, thickness uniformity, energy consumption, and service life under different combinations of process parameters, enabling the digital twin server to more effectively select the Pareto-optimal set of repair process parameters.

[0172] Furthermore, the residual of the curing kinetic equation is calculated using the relationship between curing rate and temperature:

[0173] ;

[0174] in Let j be the degree of curing at the j-th sampling point. The preset temperature and cure degree comparison table is used to compare the current temperature. The corresponding standard degree of cure, where M represents the number of sampling points along the time axis during the curing process;

[0175] The residual of the heat conduction equation is calculated using the temperature gradient:

[0176] ;

[0177] in Let J be the measured temperature change at the j-th sampling point. The predicted temperature change is calculated based on the heat balance equation ΔQ = c·m·ΔT, where ΔQ is the heat released during curing, c is the specific heat capacity of the repair material, m is the unit mass, and M represents the number of sampling points along the spatial axis.

[0178] In one embodiment of the present invention, the step of the digital twin server comparing quality data with expected indicators in the digital twin model includes:

[0179] An ultrasonic phased array detection module mounted on a robot is used to collect real-time data on the interfacial bonding quality between the repair layer and the original tube wall and upload it to a digital twin server.

[0180] The interface quality data acquired in real time by the ultrasonic phased array detection module includes the repair layer thickness value measured by the ultrasonic pulse echo method and the interface bonding strength characterization value measured by the ultrasonic transmission method.

[0181] After mapping real-time quality data to the corresponding spatial location of the digital twin model, the digital twin server generates a quality deviation distribution map. When the thickness deviation or bonding strength deviation exceeds the preset threshold, the location coordinates and range of the unqualified area are automatically marked on the digital twin model.

[0182] By equipping the repair robot with an ultrasonic phased array detection module, real-time acquisition of interfacial bonding quality data between the repair layer and the original pipe wall is achieved. The ultrasonic phased array detection module measures the thickness of the repair layer using the ultrasonic pulse-echo method and simultaneously measures the interfacial bonding strength using the ultrasonic transmission method, thus comprehensively acquiring key quality indicators of the repair layer. This data is then uploaded to a digital twin server in real time. Upon receiving this real-time quality data, the digital twin server first uses the robot's own pose information to accurately map this data to the corresponding spatial locations in the digital twin model. Subsequently, the server compares this mapped real-time quality data with the preset expected indicators in the digital twin model and generates an intuitive quality deviation distribution map. This quality deviation distribution map visually displays the distribution of repair layer thickness and interfacial bonding strength across the entire repaired pipe section. Furthermore, when the digital twin server detects that the repair layer thickness deviation or interfacial bonding strength deviation exceeds a preset threshold, the system automatically and precisely marks the location coordinates and range of the non-conforming area on the digital twin model.

[0183] In one embodiment, the specific method by which the ultrasonic phased array detection module measures the thickness of the repair layer using the ultrasonic pulse-echo method includes:

[0184] The ultrasonic phased array probe emits ultrasonic pulse waves toward the repair layer. The probe consists of multiple independent piezoelectric crystals arranged in an array. Electronic scanning and focusing of the beam are achieved by controlling the emission delay of each crystal.

[0185] The ultrasonic pulse wave propagates within the repair layer and generates a reflected echo when it encounters the interface between the repair layer and the original tube wall. The probe receives the reflected echo signal.

[0186] Measure the time difference Δt from transmission to reception of the echo, and combine it with the sound velocity V of the repair material. Calculate the thickness of the repair layer at the measurement point according to the formula h = V·Δt / 2.

[0187] The thickness distribution of the repair layer across the entire cross section is measured point by point using electronic scanning, thus forming a cross-sectional map of the repair layer thickness distribution.

[0188] The proposed solution employs an ultrasonic phased array probe, consisting of multiple independent piezoelectric crystals arranged in an array. By controlling the emission delay of each crystal, the beam can be electronically scanned and focused. This allows the probe to adjust the direction and focus of the ultrasonic beam without mechanical movement, adapting to the curved geometry of the pipe's inner wall. When the ultrasonic pulse wave is emitted and penetrates the repair layer, it generates a reflected echo upon encountering the acoustic interface between the repair layer and the original pipe wall. This reflected echo is received by the probe. By measuring the time difference Δt between ultrasonic emission and echo reception, and combining this with the predetermined sound velocity V of the repair material, the thickness of the repair layer at the current measurement point can be calculated using the formula h = V·Δt / 2.

[0189] In one embodiment, the specific method by which the ultrasonic phased array detection module measures the interfacial bonding strength characterization value using ultrasonic transmission method includes:

[0190] An ultrasonic transmitting probe is set on one side of the repair layer, and an ultrasonic receiving probe is set on the other side of the repair layer. The transmitting probe and the receiving probe are kept in a fixed relative position.

[0191] The transmitting probe emits ultrasonic pulse waves of fixed frequency and amplitude toward the repair layer. The ultrasonic pulse waves penetrate the repair layer and the interface between the repair layer and the original tube wall and are received by the receiving probe.

[0192] The amplitude attenuation ΔA and phase change Δφ of the received signal were measured. The amplitude attenuation was negatively correlated with the interface bonding quality. That is, the worse the bonding quality, the lower the ultrasonic penetration rate and the greater the amplitude attenuation.

[0193] Based on the pre-established calibration curve between bond strength and ultrasonic attenuation, the measured amplitude attenuation ΔA is converted into a characterization value of interfacial bond strength. .

[0194] By emitting ultrasonic pulses with fixed frequency and amplitude, and fixing the transmission parameters, interference from signal fluctuations at the transmitting end can be eliminated, ensuring that changes in the received signal truly originate from the inherent characteristics of the repair layer and interface, accurately reflecting the actual bonding condition. The amplitude attenuation and phase change of the received signal are measured. Utilizing the characteristic that bonding defects hinder ultrasonic propagation and increase attenuation, signal changes directly correspond to the quality of bonding, effectively distinguishing bonding areas of different qualities. Based on a pre-established calibration curve, the measured amplitude attenuation is converted into an intuitive value representing the bonding strength, transforming the acoustic signal into a quantifiable value that can be directly compared with expected indicators, facilitating rapid judgment of whether there are quality deviations exceeding thresholds. This method, combined with the overall scheme of a phased array ultrasonic detection module that collects real-time interface bonding quality data between the repair layer and the original pipe wall and uploads it to a digital twin server, enables the digital twin server to obtain accurate interface bonding strength data. The digital twin server can map real-time quality data to the corresponding spatial location in the digital twin model and generate a quality deviation distribution map. When the thickness deviation or bonding strength deviation exceeds a preset threshold, the location coordinates and range of the unqualified area are automatically marked on the digital twin model.

[0195] Furthermore, the methods for generating the quality deviation distribution map include:

[0196] The pipe segment of the digital twin model is divided into several grid units according to a preset spatial step size, and each grid unit corresponds to a detection position;

[0197] Extract the measured thickness value corresponding to each grid cell from the real-time quality data. and measured bond strength ;

[0198] Obtain the predicted thickness value for each grid cell from the repair effect prediction model. and predicted bond strength ;

[0199] Calculate the thickness deviation of each mesh cell and bond strength deviation ;

[0200] Each grid cell is colored using a color coding rule. The smaller the thickness deviation, the more green the color leans, and the larger the thickness deviation, the more red the color leans. The maximum value between the thickness deviation and the bond strength deviation is taken as the basis for coloring.

[0201] The colored mesh cells are pieced together according to their spatial positions to form a quality deviation distribution map, which is then overlaid and displayed on the 3D view of the digital twin model.

[0202] In one embodiment of the present invention, the step of adjusting the repair parameters and controlling the robot to continue working when the deviation exceeds a preset threshold includes:

[0203] Obtain the spatial distribution and deviation amplitude of non-conforming areas with deviations exceeding a preset threshold in the digital twin model;

[0204] Based on the spatial distribution of non-conforming areas, determine the locational relationship between the currently unrepaired pipe section and the non-conforming areas;

[0205] If the non-conforming area is located at the end of the repaired pipe section and adjacent to the unrepaired pipe section, the current repair parameters are used as the initial values, and the deviation amplitude of the non-conforming area is used to calibrate the repair effect prediction model online. The calibrated model is then used to re-optimize the repair process parameters of the remaining unrepaired pipe section, generate correction parameters, and issue them to the robot.

[0206] If a non-conforming area is isolated within a repaired pipe section and is not adjacent to an unrepaired pipe section, the non-conforming area will be marked as a repair point. After the entire pipe section is repaired, the robot will return to the repair point to perform local secondary repair.

[0207] In one embodiment, a specific method for online calibration of the repair effect prediction model using the deviation amplitude of the non-conforming area includes:

[0208] Obtain the actual process parameter vector of multiple measurement points within the non-conforming area. and the corresponding quality deviation amplitude ΔY;

[0209] Calculate the prediction bias vector of the current repair effect prediction model in the unqualified area:

[0210] ;

[0211] in For actual quality indicators, Predict quality metrics for the model;

[0212] The output layer weights of the repair effect prediction model are fine-tuned using the Bayesian linear regression method. The posterior distribution is calculated as follows: P(W|X, Y) ∝ P(Y|X, W)·P(W), where W is the output layer weight, P(W) is the prior distribution, and P(Y|X, W) is the likelihood function.

[0213] Only the parameters of the last two layers of the model are updated, while the parameters of the previous layers remain unchanged. The update step size is 0.001-0.01, and the number of updates is 10-50.

[0214] After calibration, the updated model is used to re-predict the quality of the remaining unrepaired pipe sections.

[0215] In one embodiment, after all repairs are completed, a digital as-built model containing the three-dimensional spatial location information of the defects, a repair process log, and quality inspection data is generated, and a repair report is output.

[0216] The digital as-built model is stored in an industrial basic class format and includes a three-dimensional geometric model of the repaired pipeline, a three-dimensional spatial location distribution map of defects, a cloud map of the actual thickness distribution of the repair layer, a cloud map of the distribution of the inverted value of the interface bond strength, pinhole and microcrack defect location markers, batch number and usage statistics of the repair materials, time history curves of key process parameters, and expected remaining service life curves based on accelerated aging tests.

[0217] The digital as-built model was imported into the city's drainage network geographic information system and linked with the historical operation and maintenance data of the pipelines, serving as the basis for future secondary repairs or pipeline upgrade decisions.

[0218] On the other hand, the present invention also provides a trenchless repair device for urban underground drainage pipes to implement the method described in any of the preceding claims, comprising:

[0219] An adaptive intelligent repair robot, and a digital twin server that communicates with the adaptive intelligent repair robot;

[0220] The repair robot includes:

[0221] The adaptive walking mechanism can adjust its own contour according to changes in the inner diameter of the pipe to maintain stable walking and measure the inner diameter of the pipe at the robot's current position;

[0222] The multi-source sensing unit integrates a structured light 3D imaging unit, a lidar, an inertial navigation system, and a distributed fiber optic acoustic sensing unit, used to collect 3D point cloud data inside the pipeline, cross-sectional point cloud data, robot pose data, and vibration response data along the pipeline.

[0223] The work execution unit connects to the robot body through a standardized interface and is used to perform at least one of the following tasks: pipeline cleaning, obstacle breaking, repair material application, or repair quality inspection.

[0224] The control unit is used to process the sensed data in real time and adjust the working parameters of the work execution unit.

[0225] The digital twin server includes: a digital twin modeling module, used to construct a three-dimensional digital model of the pipeline structure based on the perception data returned by the robot;

[0226] The repair module has a repair knowledge base and a repair effect prediction model, which is used to generate repair plans and process parameters based on defect information and environmental parameters.

[0227] The correction module is used to perform multi-objective optimization of the repair process parameters and adjust the process parameters based on real-time feedback quality data during the repair process.

[0228] In summary, after reading this invention document, those skilled in the art can make various other corresponding modifications to the technical solutions and concepts based on this invention without creative mental effort, and all of these modifications fall within the scope of protection of this invention.

Claims

1. A method for trenchless repair of urban underground drainage pipes, characterized in that, include: The robot acquires 3D point cloud data of structured light inside the pipeline, LiDAR cross-sectional data, and distributed fiber optic acoustic sensor data, and transmits them to the digital twin server. The digital twin server performs point cloud filtering and noise reduction on the collected data, and directly identifies the defect information and determines the three-dimensional spatial location of the defect through the geometric features of the defect region in the point cloud. The defect information includes the defect type and the coordinates of the defect in the point cloud coordinate system. The digital twin server uses the defect information and pipeline environment parameters to call the knowledge graph and repair effect prediction model for optimization and generate repair process parameters. The robot performs repair work according to the repair process parameters and provides real-time feedback on the quality data of the repaired layer; The digital twin server compares the quality data with the expected indicators in the digital twin model. When the deviation exceeds a preset threshold, it adjusts the repair parameters and controls the robot to continue working.

2. The method for trenchless repair of urban underground drainage pipelines according to claim 1, characterized in that, The steps for acquiring the structured light 3D point cloud data of the pipeline interior, the lidar cross-sectional data, and the distributed fiber optic acoustic sensor data collected by the robot include: An adaptive intelligent repair robot is deployed into the pipeline. The robot is equipped with a structured light 3D imaging unit, a lidar, and a distributed fiber optic acoustic sensing unit. As the robot moves along the pipeline, the structured light 3D imaging unit acquires 3D point cloud data of the inner wall of the pipeline through a laser emitter and an industrial camera, the lidar acquires point cloud data of the pipeline cross-section, and the distributed fiber optic acoustic sensing unit continuously captures vibration response data along the pipeline.

3. The method for trenchless repair of urban underground drainage pipelines according to claim 1, characterized in that, The digital twin server performs point cloud filtering and noise reduction processing on the collected data, and the steps of directly identifying defect information and determining the three-dimensional spatial location of the defect through the geometric features of the defect region in the point cloud include: The digital twin server uses a statistical filtering algorithm to remove outlier noise points from the point cloud, obtaining processed 3D point cloud data. Geometric features are extracted from the processed point cloud data to identify depth abrupt change regions, curvature abnormal regions, or continuity break regions in the point cloud as candidate defect regions. The candidate defect regions are classified and labeled to obtain the defect type and the 3D spatial coordinates of the defect in the point cloud coordinate system.

4. The method for trenchless repair of urban underground drainage pipelines according to claim 1, characterized in that, The steps of the digital twin server processing the collected data and determining the three-dimensional spatial location of the defect also include: The robot's 3D pose is associated with the point cloud coordinate system: the robot's 3D position and attitude data are obtained in real time from the inertial navigation system on the robot, and a global coordinate system is established based on the positioning data of the inertial navigation system; the 3D spatial coordinates of the defect in the point cloud coordinate system are mapped to the global coordinate system through coordinate transformation to obtain the global 3D spatial position information of the defect.

5. The method for trenchless repair of urban underground drainage pipelines according to claim 4, characterized in that, The association between the robot's three-dimensional pose and the point cloud coordinate system includes: The point cloud density and feature point distribution uniformity of the lidar point cloud data are obtained, and the localization contribution of the point cloud data is evaluated based on the above indicators. The gyroscope drift rate, accelerometer noise level, and odometry cumulative error of the inertial navigation system are obtained, and the instantaneous positioning accuracy of the inertial navigation system is evaluated based on the above indicators. An energy-sensing adaptive fusion method is adopted to dynamically allocate the fusion weights of point cloud data and inertial navigation data according to the complexity of the pipeline environment and the robot's motion state; The relative pose obtained by point cloud registration and the absolute pose calculated by inertial navigation are fused based on the fusion weight to obtain the robot's three-dimensional pose in the global coordinate system. The point cloud observation value and inertial navigation output value at the next moment are predicted based on the 3D pose and compared with the actual collected data to obtain the fusion residual. When the fusion residual exceeds the preset threshold, the fusion weight of the corresponding data source is reduced.

6. The method for trenchless repair of urban underground drainage pipelines according to claim 4, characterized in that, The steps for obtaining the robot's three-dimensional position and attitude data output in real time by the robot's inertial navigation system include: Using miniaturized gyroscopes and accelerometers as inertial measurement units, the robot's three-dimensional coordinates are measured directly inside the tube. When the robot travels inside the pipeline to the location of the inspection well with known coordinates on the ground, the cumulative error of the inertial navigation system is corrected at zero speed or calibrated using ground differential GPS coordinates. An error compensation algorithm is used to compensate for the drift error of the inertial navigation system in real time.

7. The method for trenchless repair of urban underground drainage pipelines according to claim 1, characterized in that, The digital twin server, based on defect information and pipeline environment parameters, calls a knowledge graph and a repair effect prediction model for optimization, and generates repair process parameters through the following steps: The digital twin server calls a pre-built knowledge graph of repair solutions, taking the defect type, the three-dimensional spatial location of the defect, the pipe material parameters, and the construction site environmental data as input, and calculates candidate repair solutions through a graph neural network. For candidate repair schemes, a repair effect prediction model based on neural networks is invoked for multi-objective optimization. The objective functions of the multi-objective optimization include maximizing the bonding strength of the repair layer, minimizing the thickness non-uniformity coefficient of the repair layer, and minimizing the repair energy consumption and material cost. A non-dominated sorting genetic algorithm with an elitist strategy is used for iterative solution to obtain the Pareto optimal set of repair process parameters.

8. The method for trenchless repair of urban underground drainage pipelines according to claim 1, characterized in that, The steps by which the digital twin server compares quality data with expected metrics in the digital twin model include: An ultrasonic phased array detection module mounted on a robot is used to collect real-time data on the interfacial bonding quality between the repair layer and the original tube wall and upload it to a digital twin server. The interface quality data acquired in real time by the ultrasonic phased array detection module includes the repair layer thickness value measured by the ultrasonic pulse echo method and the interface bonding strength characterization value measured by the ultrasonic transmission method. After mapping real-time quality data to the corresponding spatial location of the digital twin model, the digital twin server generates a quality deviation distribution map. When the thickness deviation or bonding strength deviation exceeds the preset threshold, the location coordinates and range of the unqualified area are automatically marked on the digital twin model.

9. The method for trenchless repair of urban underground drainage pipelines according to claim 1, characterized in that, The steps of adjusting the repair parameters and controlling the robot to continue working when the deviation exceeds a preset threshold include: Obtain the spatial distribution and deviation amplitude of non-conforming areas with deviations exceeding a preset threshold in the digital twin model; Based on the spatial distribution of non-conforming areas, determine the locational relationship between the currently unrepaired pipe section and the non-conforming areas; If the non-conforming area is located at the end of the repaired pipe section and adjacent to the unrepaired pipe section, the current repair parameters are used as the initial values, and the deviation amplitude of the non-conforming area is used to calibrate the repair effect prediction model online. The calibrated model is then used to re-optimize the repair process parameters of the remaining unrepaired pipe section, generate correction parameters, and issue them to the robot. If a non-conforming area is isolated within a repaired pipe section and is not adjacent to an unrepaired pipe section, the non-conforming area will be marked as a repair point. After the entire pipe section is repaired, the robot will return to the repair point to perform local secondary repair.

10. A trenchless repair device for urban underground drainage pipelines to implement the method of any one of claims 1-9, characterized in that, include: An adaptive intelligent repair robot, and a digital twin server that communicates with the adaptive intelligent repair robot; The repair robot includes: The adaptive walking mechanism can adjust its own contour according to changes in the inner diameter of the pipe to maintain stable movement; The multi-source sensing unit integrates a structured light 3D imaging unit, a lidar, an inertial navigation system, and a distributed fiber optic acoustic sensing unit, used to collect 3D point cloud data inside the pipeline, cross-sectional point cloud data, robot pose data, and vibration response data along the pipeline. The work execution unit connects to the robot body through a standardized interface and is used to perform at least one of the following tasks: pipeline cleaning, obstacle breaking, repair material application, or repair quality inspection. The control unit is used to process the sensed data in real time and adjust the working parameters of the work execution unit. The digital twin server includes: a digital twin modeling module, used to construct a three-dimensional digital model of the pipeline structure based on the perception data returned by the robot; The repair module has a repair knowledge base and a repair effect prediction model, which is used to generate repair plans and process parameters based on defect information and environmental parameters. The correction module is used to perform multi-objective optimization of the repair process parameters and adjust the process parameters based on real-time feedback quality data during the repair process.

Images