Drainage pipeline inspection control system based on adaptive path planning

Abstract

The application discloses a drainage pipeline inspection control system based on adaptive path planning and particularly relates to the technical field of pipeline detection and automatic control, and comprises the following steps: acquiring multi-dimensional environment information data inside a target drainage pipeline; constructing a pipeline passability probability map based on the multi-dimensional environment information data; receiving the pipeline passability probability map and a preset inspection task instruction, and generating an inspection path which can be dynamically adjusted; controlling an inspection robot to move along the planned path, and comparing the planned expectation with the actual execution effect in real time. Through a multi-modal environment perception module, a pipeline passability dynamic modeling module, a multi-target adaptive path planning engine module and a closed-loop execution and learning module, the application constructs a drainage pipeline inspection control system based on adaptive path planning, and solves the problems of insufficient data fusion capability, lack of quantitative model for passability evaluation, weak dynamic map updating capability and lack of closed-loop execution and self-learning mechanism in the prior art.

Description

Technical Field

[0001] This invention relates to the field of pipeline inspection and automatic control technology, and more specifically, to a drainage pipeline inspection and control system based on adaptive path planning. Background Technology

[0002] As a core component of urban infrastructure, drainage pipelines play a crucial role in rainwater drainage and sewage transport. Their operational status directly affects urban flood control and drainage capabilities, water quality, and public safety. With the acceleration of urbanization, combined sewer systems (with diameters ranging from DN300 to DN2000) have long faced harsh operating conditions such as low sunlight, high dust levels, accumulation of toxic and harmful gases (such as hydrogen sulfide and methane), silt buildup, and complex water flow. Defects such as pipe cracks, corrosion, damage, and misaligned joints occur frequently. If these defects are not inspected and repaired in a timely manner, they can easily lead to safety accidents such as pipe leaks, collapses, and sewage overflows. Therefore, regular pipeline inspections are of paramount importance.

[0003] To address the pain points of manual inspection, robot-assisted inspection technology has been gradually applied. However, existing robot inspection systems still have many technical shortcomings in practical applications, making it difficult to meet the high-precision and high-efficiency inspection requirements in complex pipeline environments. The environmental perception dimension is limited and the data fusion capability is insufficient: existing inspection robots mostly use a single sensor (such as a single camera or lidar) to collect environmental information, which fails to comprehensively cover multi-dimensional data such as pipe wall geometry, silt characteristics, water flow status, harmful gas concentration and robot body status. The lack of a quantitative model for accessibility assessment and the weak ability to dynamically update maps: The existing system's judgment on the accessibility of pipelines is mostly based on the simple determination of whether there are obstacles. It has not established a multi-factor comprehensive quantitative assessment system for access costs and ignores key influencing factors such as the thickness of silt, the impact force of water flow, and the proximity of defect areas. Path planning algorithms have significant limitations and poor multi-objective optimization performance: existing robot path planning mostly adopts traditional A / B algorithm. Basic search algorithms such as the Dijkstra algorithm only take "shortest distance" as the optimization goal, without taking into account multiple dimensions such as total inspection time, defect detection confidence, robot energy consumption, and pipeline cross-section coverage. The lack of closed-loop execution and self-learning mechanisms results in weak system adaptability: the existing system's "planning-execution" process is in an open-loop state, and there is no real-time comparison mechanism between the planned expectations and the actual execution results. The response to robot position and posture deviations, environmental parameter fluctuations, and other situations is lagging, making it difficult to avoid execution errors through dynamic adjustments.

[0004] In summary, existing drainage pipeline inspection technologies are insufficient in terms of the comprehensiveness of environmental perception, the accuracy of accessibility assessment, the multi-objective optimization capability of path planning, and the adaptability of the system. There is an urgent need to develop an inspection control system that can adapt to complex and dynamic pipeline environments, take into account multi-dimensional inspection objectives, and has continuous optimization capabilities. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, the present invention provides a drainage pipeline inspection and control system based on adaptive path planning, which solves the problems mentioned in the background art through the following scheme.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a drainage pipeline inspection and control system based on adaptive path planning, comprising: Multimodal environmental perception module: acquires multidimensional environmental information data inside the target drainage pipe, including the geometric shape and defect features of the pipe wall, the three-dimensional contour features of the silt, the water flow speed and direction, the distribution of harmful gas concentration, and the robot's body state; Pipeline navigability dynamic modeling module: Based on the multi-dimensional environmental information data, a pipeline navigability probability map is constructed. The map contains spatial obstacle information and marks the navigability cost for each navigable area. The navigability cost is calculated by comprehensively considering the thickness of silt, water flow impact force, and proximity to suspected defect areas. Multi-objective adaptive path planning engine module: The engine receives the pipeline drivability probability map and preset inspection task instructions, and takes the total inspection time, defect detection confidence, robot energy consumption and full coverage of pipeline cross section as optimization objectives. By integrating the improved search algorithm and model prediction control algorithm, it generates a dynamically adjustable inspection path. Closed-loop execution and learning module: controls the inspection robot to move along the planned path and compares the planned expectations with the actual execution results in real time. If the deviation exceeds the threshold or a sudden obstacle is detected, the inspection path is adjusted immediately. If the inspection is successfully completed, the inspection data will be used as the passage cost parameter of the pipeline accessibility probability map for system updates.

[0007] The technical effects and advantages of this invention are as follows: 1. This invention adopts a multi-modal sensor array integrated design, which includes a 16-line three-dimensional LiDAR, a 4K industrial camera, a multi-point ultrasonic flow meter, a multi-gas sensor module and a robot body status sensor, to achieve full-dimensional data acquisition of pipe wall features, sludge contours, water flow status, harmful gas concentrations and robot posture, thus solving the pain points of existing technologies such as single perception dimension and insufficient data fusion. 2. This invention refines the pipeline space through three-dimensional raster modeling, establishes quantitative judgment thresholds for multiple types of obstacles (silt, pipe wall defects, fluid, gas), and constructs a multi-factor passage cost assessment system based on "silt thickness + water flow impact force + proximity of defect area". 3. This invention constructs a multi-objective optimization function with the objectives of "shortest total inspection time, highest defect detection confidence, lowest robot energy consumption, and largest cross-sectional coverage," and uses the Analytic Hierarchy Process (AHP) to dynamically prioritize the objectives; it also incorporates and improves the AHP algorithm. The algorithm and model predictive control algorithm first completes the global path search, and then achieves dynamic adjustment through local rolling optimization; 4. This invention constructs a closed-loop process of "planning-execution-monitoring-adjustment-update", which monitors the robot's position and posture, environmental parameters and execution effect in real time, and sets graded deviation thresholds (slight, moderate and severe) to correspond to different adjustment strategies. After the inspection is completed, the path planning algorithm parameters are optimized through data validity verification, and a knowledge database of working condition-strategy mapping is established. Attached Figure Description

[0008] Figure 1 This is a schematic diagram of the system operation logic structure of the present invention.

[0009] Figure 2 This is a schematic diagram of the multimodal environment perception module structure of the present invention.

[0010] Figure 3 This is a schematic diagram of the multi-objective adaptive path planning engine module structure of the present invention.

[0011] Figure 4 This is a schematic diagram of the closed-loop execution and learning module structure of the present invention. Detailed Implementation

[0012] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] Please see Figure 1 - Figure 4 As shown, this embodiment of the invention provides a drainage pipeline inspection and control system based on adaptive path planning, including: Multimodal environmental perception module: acquires multidimensional environmental information data inside the target drainage pipe, including the geometric shape and defect features of the pipe wall, the three-dimensional contour features of the silt, the water flow speed and direction, the distribution of harmful gas concentration, and the robot's body state; Pipeline navigability dynamic modeling module: Based on the multi-dimensional environmental information data, a pipeline navigability probability map is constructed. The map contains spatial obstacle information and marks the navigability cost for each navigable area. The navigability cost is calculated by comprehensively considering the thickness of silt, water flow impact force, and proximity to suspected defect areas. Multi-objective adaptive path planning engine module: The engine receives the pipeline drivability probability map and preset inspection task instructions, and takes the total inspection time, defect detection confidence, robot energy consumption and full coverage of pipeline cross section as optimization objectives. By integrating the improved search algorithm and model prediction control algorithm, it generates a dynamically adjustable inspection path. Closed-loop execution and learning module: controls the inspection robot to move along the planned path and compares the planned expectations with the actual execution results in real time. If the deviation exceeds the threshold or a sudden obstacle is detected, the inspection path is adjusted immediately. If the inspection is successfully completed, the inspection data will be used as the passage cost parameter of the pipeline accessibility probability map for system updates.

[0014] The multimodal environment perception module's core function is to collect multidimensional environmental information data and robot body state data from inside the drainage pipe, and then preprocess and fuse these data to form a standardized dataset, providing data support for subsequent modules. The specific implementation is as follows: A101: Sensor Array Selection, Deployment, and Calibration A1011: Sensor selection is for combined sewer systems with pipe diameters of DN300-DN2000, suitable for operating conditions such as low light, high dust, and presence of toxic and harmful gases within the pipes. A multi-modal sensor array is configured, and the selection parameters for each sensor are as follows: 3D LiDAR: It adopts a 16-line LiDAR with a ranging accuracy of ±2cm, a scanning frequency of 10Hz, and has the ability to resist strong light interference. It can work stably in an environment with humidity from 0 to 100%. High-definition industrial camera: Utilizes a 4K resolution industrial camera with a frame rate of 25fps, equipped with an anti-fog and waterproof lens and an infrared fill light, supports HDR mode, and has a lens distortion rate controlled within 1%; Ultrasonic flow meter: A multi-point measurement flow meter is selected, with a measurement range of 0.01-10m / s and an accuracy of ±1%. It supports the simultaneous acquisition of flow velocity data at three measurement points at the top, middle, and bottom of the pipeline. Multi-gas sensor module: integrates hydrogen sulfide, methane, carbon monoxide and oxygen detection units, with detection accuracy controlled within ±5%FS, response time ≤3s, and over-limit alarm function; Robot body status sensors: GPS navigation with a positioning accuracy of ±3cm; six-axis attitude sensor with a measurement range of ±180° and an accuracy of ±0.5°; and a power sensor with a sampling frequency of 1Hz.

[0015] A1012: Sensor deployment. The aforementioned sensor array will be integrated into the sensing cabin of the inspection robot. The specific installation location is as follows: The 3D LiDAR and high-definition industrial camera are installed at the front of the robot. The LiDAR's scanning field of view covers the entire cross section of the pipe, and the camera lens is parallel to the LiDAR scanning plane to ensure that the image data and point cloud data are spatially aligned. The ultrasonic flow meter is installed on both sides of the robot, with three measuring points corresponding to the top, middle and bottom of the pipe cross section, respectively, and the distance between the measuring points is 1 / 3 of the pipe diameter; The multi-gas sensor module is installed on the top of the robot, with the air inlet facing the direction of the airflow in the pipeline, to avoid interference from the robot's own exhaust on the detection results; The robot's body status sensors are embedded inside the control cabin, and the mounting reference planes of the positioning module and attitude sensors are parallel to the robot chassis.

[0016] A1013: Sensor calibration. After completing sensor installation, perform accuracy calibration and time synchronization. The specific steps are as follows: LiDAR calibration: The planar calibration method is adopted. Using a standard calibration board as a reference, the calibration board is placed at different positions within the scanning range of the LiDAR, and the point cloud data of the calibration board is collected to correct the angle error and distance error of the LiDAR, ensuring that the planar fitting error of the point cloud data after calibration is ≤0.5mm. Industrial camera calibration: Zhang Zhengyou calibration method is adopted. By shooting the checkerboard calibration board at different angles, the intrinsic parameter matrix and distortion coefficient of the camera are calculated, lens distortion is eliminated, and the mapping relationship between pixel coordinates and world coordinates is established. Ultrasonic flow meter calibration: Place the flow meter in a standard flow rate water tank and collect data at three standard flow rates of 0.5 m / s, 1.0 m / s, and 1.5 m / s. Record the measurement error and generate a compensation coefficient table. After calibration, the measurement error should be ≤0.5%. Time synchronization calibration: Connect all sensors to the robot's PTP precision time protocol module and set the synchronization accuracy to ≤1ms to ensure that the timestamps of the data collected by each sensor are consistent.

[0017] A102: Multi-dimensional data acquisition with time synchronization A1021: Data acquisition triggering mechanism settings, adopting a dual acquisition mechanism of distance triggering + time triggering, with specific parameters as follows: Distance trigger: Every time the robot travels 0.5m, it triggers a full-section scan of the LiDAR and industrial camera to ensure that there is at least one complete set of point cloud data and image data within every 0.5m length of the pipeline; Time-triggered: Every 1 second, the ultrasonic flow meter, multi-gas sensor module and body status sensor are triggered to collect data to ensure that the robot can still acquire continuous environmental and status data when it is stationary or moving at low speed.

[0018] A1022: Data acquisition is performed by category. According to the triggering mechanism, five types of data are collected sequentially: pipe wall geometry and defect features, three-dimensional contour features of sediment, water flow velocity and direction, distribution of harmful gas concentration, and robot body state. The specific operation is as follows: Pipe wall geometry and defect feature acquisition: The lidar emits a laser beam to scan the inner wall of the pipe, generating 3D data containing millions of point clouds, and recording the spatial coordinates of each point. With reflection intensity; industrial cameras simultaneously capture images of the pipe wall, each image corresponding to one scanning cycle of the LiDAR, and image annotation tools are used to mark the location coordinates and morphological parameters of defects such as cracks, corrosion, damage, and misalignment of interfaces in the drainage pipe; 3D contour feature acquisition of sediment: The point cloud data collected by lidar is segmented, and the pipe wall point cloud is filtered out. The remaining point cloud is the sediment point cloud. The 3D contour of the sediment is extracted by clustering algorithm, and parameters such as the maximum thickness, distribution range, and surface flatness of the sediment are recorded. Water flow velocity and direction acquisition: The ultrasonic flow meter emits ultrasonic signals at three measuring points simultaneously. The flow velocity and direction at each measuring point are calculated based on the time difference of signal propagation. The average flow velocity and main flow direction of the pipe cross section are obtained through vector synthesis algorithm, and the turbulence intensity of the water flow is recorded at the same time. Harmful gas concentration distribution acquisition: The multi-gas sensor module acquires data at a frequency of 1 second / time. Combined with the robot's positioning information, a concentration distribution curve is generated with the pipe length as the horizontal axis and the gas concentration as the vertical axis, marking the location and range of areas with excessive concentration. Robot body status acquisition: The positioning module outputs the robot's longitude, latitude, and elevation in real time; the attitude sensor records the robot's pitch angle, roll angle, and heading angle; the power sensor collects the remaining battery power, voltage, and current; the motor load sensor monitors the torque and speed of the drive wheels to determine whether the robot is slipping or overloaded.

[0019] A103: Multi-source data preprocessing and fusion A1031: Single-source data preprocessing involves targeted preprocessing of the raw data collected by each sensor. The specific operations are as follows: LiDAR point cloud preprocessing: A combination of statistical filtering and radius filtering algorithms is used for noise reduction. The number of neighboring points for statistical filtering is set to 5, and the standard deviation threshold is set to 1.0. Isolated noise points that deviate from the point cloud cluster are removed. The point cloud was divided into 0.5cm×0.5cm×0.5cm voxel grids using a voxel grid method, with each grid retaining a center point; the point cloud data in the lidar coordinate system was converted to data in the robot coordinate system through coordinate transformation. Industrial camera image preprocessing: The original image is sequentially processed by grayscale conversion, histogram equalization, Gaussian filtering, and edge detection. Grayscale conversion reduces the amount of computation, histogram equalization improves image contrast, Gaussian filtering uses a 5×5 filter kernel with a standard deviation of 1.5 to eliminate image noise, and the Canny edge detection algorithm is used to extract the edge contours of defects. Fluid and gas data preprocessing: Multi-point velocity data acquired by the ultrasonic flow meter are processed using... The criteria filter out outliers. A moving average filter is used on the concentration data collected by multiple gas sensors, with a window size of 5, to eliminate instantaneous fluctuations. Body state data preprocessing: Kalman filtering is applied to the positioning data, GPS and inertial navigation data are fused, and positioning deviations when GPS signals are lost are corrected; complementary filtering is applied to the attitude data to eliminate angle errors caused by sensor drift.

[0020] A1032: Multi-source data spatiotemporal fusion is based on PTP time synchronization, outputting a unified timestamp and spatial coordinates of the robot coordinate system to construct a three-dimensional data model of time, space, and attributes; it associates LiDAR point cloud data, camera image data, flow velocity data, gas concentration data, and body state data under the same timestamp and spatial location to generate a standardized multi-dimensional environmental information dataset.

[0021] The pipeline accessibility dynamic modeling module, at its core, constructs a high-precision, dynamically updated pipeline accessibility probability map based on the fused multi-dimensional environmental information dataset. This provides a quantitative environmental assessment basis for route planning, specifically including: B201: Pipeline Space Rasterized Refined Modeling B2011: Grid division parameter settings. Based on the pipe diameter and sensor accuracy, determine the basic unit size for gridding. The formula for calculating the grid side length is as follows: ,in The diameter of the pipe; For example: For a DN1000 pipe, the grid side length is set to 10cm, and the grid is positioned along the pipe extension direction using the robot coordinate system as a reference. Axis, Pipe Radial Axis, Pipe Circumferential The internal space of the pipe is divided into a three-dimensional grid array along three axes, with each grid assigned a unique spatial index coordinate. .

[0022] B2012: Raster Attribute Initialization configures basic attribute fields for each raster, including spatial coordinates, initial access status, initial access cost, and update timestamp. The initial access status defaults to undetected, the initial access cost defaults to 0, and the update timestamp defaults to the time when the sensor collected the data. A raster attribute database is established, supporting real-time querying and modification.

[0023] B202: Multi-dimensional obstacle information identification and hierarchical labeling B2021: Obstacle assessment criteria are established based on a multi-dimensional environmental information dataset, setting thresholds for different types of obstacles. Specifically, this includes: Silt obstruction: When the thickness of silt within the grid... When the debris reaches 8 cm, it is determined to be an obstruction. The robot's chassis height is 10 cm, with a 2 cm safety margin. Pipe wall defect obstruction: When the depth of the pipe wall defect within the grid A defect is defined as a defect if it is greater than 3 mm or if the length of the defect exceeds 50% of the grid side length. Fluid obstruction: Water flow velocity within the grid Greater than At that time, or when the angle between the water flow direction and the robot's direction of travel is greater than 100°C. At that time, it was determined to be a fluid obstruction; Gas Hazard: Hydrogen sulfide concentration within the grid A value greater than 10 ppm is considered a gas obstruction.

[0024] B2022: Raster traffic status hierarchical labeling Based on obstacle type, grid passage status is divided into four levels, with specific labeling rules including: Impassable grid: If there is any one of the following obstacles: siltation, pipe wall defects, or both, or other obstacles, it is marked in red. Restricted access grid: There is a fluid or gas obstacle, but the degree of obstacle does not reach the standard of being impassable, and it is marked in yellow; Free passage grid: No obstacles, all environmental parameters meet safety standards, marked in green; Undetected grid: Areas where no valid environmental data was collected, marked in gray; Obstacles are identified for each grid cell, and the passage status results are written into the grid attribute database, with the obstacle type and judgment criteria labeled.

[0025] B203: Multi-factor Quantitative Calculation and Normalization of Passage Cost Traffic cost factor extraction and normalization, selecting sediment thickness Water flow impact force Proximity of defect areas The three core factors are mapped to the [0,1] interval using a min-max normalization algorithm to eliminate dimensional differences. The specific formulas include: Normalized value of sediment thickness: ,in , ; Normalized value of water flow impact force: ,in The density of the wastewater is taken as 1050 kg / m³. 3 , The cross-sectional area of ​​the grid is... , ; Defect region proximity normalized value: ,in The distance from the current grid cell to the nearest defect region. , ; The weighted calculation of passage cost uses a weighted linear combination model to calculate the final passage cost of each grid cell. The specific calculation formula is as follows: in Let be the weighting coefficient, satisfying When the inspection task priority is defect detection priority, set When the task priority is set to fast inspection first, set... After the calculation is completed, Write to the raster attribute database. The value ranges from [0,1], with larger values ​​indicating greater difficulty in passage.

[0026] B204: Generation and Dynamic Updating of Accessibility Probability Maps The probability map visualization is built based on a raster attribute database and uses a combination of color coding and numerical labeling to generate a probability map of pipeline accessibility. The map is divided into two layers: the status layer uses different colors to mark the grid passage status; the cost layer marks the passage cost value in each grid. At the same time, key information such as defect areas, areas with excessive gas, and high flow rate areas are marked on the map to generate a visual interface.

[0027] The map dynamic update mechanism is set up, and map update trigger conditions are established, including distance trigger (every time the robot moves 1m), time trigger (every 10s), and event trigger (when a sudden obstacle is detected).

[0028] When the triggering conditions are met, the latest multi-dimensional environmental information data is automatically extracted, the grid access status and access cost of the corresponding area are updated, and the update timestamp is recorded. After the update is completed, a map version log is generated to save historical version data, supporting backtracking and comparative analysis.

[0029] The multi-objective adaptive path planning engine module, at its core, generates dynamic inspection paths that meet multi-objective optimization requirements based on a drivability probability map. The specific implementation is as follows: C301: Detailed Analysis and Prioritization of Inspection Task Instructions The task instruction parameter extraction receives preset inspection task instructions and parses out the core parameter set, including: a list of coordinates of mandatory inspection areas. Requirements for grid coverage of the entire cross-section of the pipeline Total inspection time limit Lower limit of confidence level for defect detection Robot energy consumption limit Time step of path planning .

[0030] Task priorities are dynamically sorted. The Analytic Hierarchy Process (AHP) is used to prioritize four optimization objectives: total inspection time, defect detection confidence, robot energy consumption, and cross-sectional coverage. A judgment matrix is ​​constructed, and the weight value of each objective is calculated.

[0031] C302: Construction of Multi-Objective Optimization Functions and Setting of Constraints A multi-objective optimization function is modeled and constructed with the objectives of "shortest total inspection time, highest defect detection confidence, lowest robot energy consumption, and maximum cross-sectional coverage." The specific calculations include: in For the number of path nodes, For the travel time between nodes, For the number of defect detection points, For the first The detection confidence level of each defect For robot standby power, For robot travel power, For the distance traveled between nodes, For the number of grid cells covered, This represents the total number of grid cells.

[0032] Constraints are set to clearly define the boundaries of path planning, including: path nodes must not fall into impassable grid cells; the total cost of path travel. The robot's remaining battery power is ≥20%; the dwell time at the defect detection point is ≥2s.

[0033] C303: Planning for the Integration of Improved Search Algorithm and Model Predictive Control Algorithm Improvement A The algorithm's global path planning optimizes the heuristic function of the traditional A algorithm, making the heuristic function... Improved to ; in The total cost of travel from the current node to the target node. The cost weighting coefficient addresses the problem of the traditional A algorithm, which only considers distance and ignores environmental costs.

[0034] Using the robot's starting point as the initial node, the task's ending point as the target node, and a drivability probability map as the search space, an improved A algorithm is used for path search to generate a sequence of key nodes for the globally optimal path. Each node contains information such as coordinates, passage cost, estimated travel time, and estimated energy consumption.

[0035] The Model Predictive Control (MPC) algorithm is used for local dynamic adjustment. A robot kinematics model is established, with the robot's velocity and steering angle as inputs, and the output being the next five sampling periods. The system predicts the location and attitude of the road within the area; simultaneously, it establishes an environmental prediction model and, based on historical updates of the accessibility probability map, predicts the trend of road cost changes over the next five sampling periods.

[0036] Using the key node sequence of the global path as a reference trajectory, the robot's current position, posture, and environmental data are collected in each sampling period, and the deviation between the actual and predicted values ​​is calculated. Using a multi-objective optimization function as the objective and robot kinematic constraints as the boundary, the optimal control quantities (velocity, steering angle) are solved through rolling optimization to fine-tune the local path. If the path still cannot meet the constraints after fine-tuning, local path replanning is triggered.

[0037] C304: Path Feasibility Verification and Iterative Optimization Route feasibility verification verifies the planned route from four dimensions: accessibility, timeliness, energy consumption, and coverage. It checks whether the route passes through impassable grids, calculates whether the total inspection time is ≤120min, calculates whether the estimated total energy consumption is ≤500Wh, and calculates whether the cross-sectional coverage is ≥95%.

[0038] Iterative optimization: if the verification passes, output the final dynamic inspection path; if the verification fails, adjust the weight coefficients. and Then, the global path planning and local adjustment process is repeated until a feasible path that meets all constraints is generated.

[0039] The closed-loop execution and learning module, at its core, controls the robot to precisely execute the inspection path and achieves self-learning and iterative optimization of the system through execution feedback. Specific implementation includes: D401: Path Execution and Real-time Monitoring of Multi-Dimensional Status The path execution command is issued, converting the dynamic inspection path into robot motion control commands, including travel speed (0.1-0.5 m / s) and turning angle. Parameters such as dwell time ≥2s are sent to the robot via the CAN bus. The drive uses a PID control algorithm to control the speed and steering angle of the robot's drive wheels, ensuring that the robot moves along the planned path.

[0040] Multi-dimensional status monitoring: Activate the robot's status monitoring unit to collect and record the following monitoring data in real time: Position and attitude monitoring: Real-time acquisition of the robot's actual position through positioning and attitude sensors. pitch angle in attitude Roll angle Heading angle ; Environmental parameter monitoring: Through the multimodal sensing module, environmental parameters such as the thickness of silt, water flow speed, and gas concentration at the robot's current location are acquired in real time; Execution performance monitoring: Record the robot's actual travel time, energy consumption, defect detection confidence level, and other execution parameters.

[0041] The deviation calculation between planning and actual values, based on the same timestamp, calculates the deviation between the planned and actual values, including: location deviation: Among them, attitude deviation , , Environmental parameter deviation , ; Execution parameter deviation , .

[0042] D402: Deviation Assessment and Graded Emergency Response Adjustment Based on the robot's control precision and inspection task requirements, different types of deviation warning thresholds and emergency thresholds are set, including: Position deviation warning threshold Emergency threshold Attitude deviation warning threshold Emergency threshold .

[0043] The tiered adjustment strategy is implemented, and minor deviations are adjusted (deviation < warning threshold): the robot maintains its original path and fine-tunes the speed of the drive wheels and the steering angle through the PID control algorithm to gradually eliminate the deviation; Moderate deviation adjustment (warning threshold ≤ deviation < emergency threshold): Pause robot movement, call the model predictive control algorithm, replan the local path based on the latest environmental data and deviation value, adjust the path node coordinates, and continue moving after the replanning is completed; Severe deviation adjustment (deviation ≥ emergency threshold or sudden obstacle detected): immediately activate emergency braking to stop the robot's movement; trigger the multimodal perception module to perform a full-section scan and update the accessibility probability map; Calling Improved A The algorithm replans the global path to avoid unexpected obstacles; after the new path is generated, the robot is controlled to move along the new path until it returns to the key node of the original planned path or completes the inspection task.

[0044] D403: Inspection Data Verification and Model Parameter Iterative Update D4031: Inspection Data Validity Verification After completing all inspection tasks, activate the data verification unit and verify the data validity using the following methods: The criteria include removing outliers, verifying data integrity, checking for missing data in mandatory inspection areas, comparing defect annotation results between LiDAR data and camera image data, and verifying data consistency.

[0045] D4032: Toll Cost Model Parameter Update Extract the actual values ​​of silt thickness, water flow impact force, and proximity of defect areas from the valid data, and calculate the error between the actual values ​​and the model calculation values. Use the least squares method to fit the error data and correct the weighting coefficients in the toll cost model. This makes the model's calculated values ​​more closely resemble the actual environment.

[0046] D4033: Path Planning Algorithm Parameter Optimization Extract parameters such as path execution time, energy consumption, and coverage from the valid data, and calculate the error between these parameters and the planned values; based on the error data, adjust and improve A. heuristic function weight coefficients of the algorithm Adjusting the sampling period of the model predictive control algorithm improves the planning accuracy of the algorithm.

[0047] D4034: Iterative Optimization of Accessibility Probability Map The valid data from this inspection is written into the grid attribute database, and the access status and access cost of the corresponding grid are updated. Based on historical inspection data, the random forest algorithm is used to predict the changing trend of the pipeline environment, generate a trend prediction layer, and overlay it onto the accessibility probability map.

[0048] D404: System Self-Learning Iteration and Knowledge Accumulation The updated passage cost model parameters and path planning algorithm parameters are written into the system configuration file. When the next inspection task is started, the multi-objective adaptive path planning engine module will directly call the updated parameters to generate a better inspection path. At the same time, an inspection knowledge database is established to record the optimal weight coefficient combination, path planning strategy, and emergency adjustment plan under different pipeline operating conditions, forming a standardized operating condition-strategy mapping table to realize the system's self-learning and intelligent upgrade.

[0049] Secondly: The accompanying drawings of the embodiments disclosed in this invention only involve the structures involved in the embodiments disclosed in this invention. Other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of this invention can be combined with each other. In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A drainage pipeline inspection and control system based on adaptive path planning, characterized in that, include: Multimodal environmental perception module: acquires multidimensional environmental information data inside the target drainage pipe, including the geometric shape and defect features of the pipe wall, the three-dimensional contour features of the silt, the water flow speed and direction, the distribution of harmful gas concentration, and the robot's body state; Pipeline navigability dynamic modeling module: Based on the multi-dimensional environmental information data, a pipeline navigability probability map is constructed. The map contains spatial obstacle information and marks the navigability cost for each navigable area. The navigability cost is calculated by comprehensively considering the thickness of silt, water flow impact force, and proximity to suspected defect areas. Multi-objective adaptive path planning engine module: The engine receives the pipeline drivability probability map and preset inspection task instructions, and takes the total inspection time, defect detection confidence, robot energy consumption and full coverage of pipeline cross section as optimization objectives. By integrating the improved search algorithm and model prediction control algorithm, it generates a dynamically adjustable inspection path. Closed-loop execution and learning module: controls the inspection robot to move along the planned path and compares the planned expectations with the actual execution results in real time: if the deviation exceeds the threshold or a sudden obstacle is detected, the inspection path is adjusted immediately; If the inspection is successfully completed, the inspection data will be used as the passage cost parameter in the pipeline accessibility probability map for system updates.

2. The drainage pipeline inspection and control system based on adaptive path planning according to claim 1, characterized in that, The multimodal environment perception module includes: A multimodal sensor array is integrated on the inspection robot, including a 3D LiDAR, an industrial camera, an ultrasonic flow meter, a multi-gas sensor module, and robot body status sensors. The data acquisition unit is configured to use a mechanism that combines distance triggering and time triggering to synchronously acquire the multidimensional environmental information data; The data fusion unit is configured to preprocess the collected raw data and perform spatiotemporal fusion based on a unified timestamp and spatial coordinates to generate a standardized multidimensional environmental information dataset.

3. The drainage pipeline inspection and control system based on adaptive path planning according to claim 2, characterized in that, The selection parameters for the multimodal sensor array include: The 3D lidar has 16 lines, a ranging accuracy of ±2cm, a scanning frequency of 10Hz, and is capable of resisting strong light interference and can work stably in environments with humidity from 0 to 100%. The high-definition industrial camera features 4K resolution, a frame rate of 25fps, an anti-fog and waterproof lens, an infrared fill light, supports HDR mode, and has a lens distortion rate of ≤1%. The ultrasonic flow meter is a multi-point measurement type, with a measurement range of 0.01-10 m / s and an accuracy of ±1%. The multi-gas sensor module integrates hydrogen sulfide, methane, carbon monoxide, and oxygen detection units, with a detection accuracy of ±5%FS, a response time of ≤3s, and an alarm function for exceeding the limit. The robot's body state sensors include a GPS navigation module with a positioning accuracy of ±3cm, a six-axis attitude sensor with a measurement range of ±180° and an accuracy of ±0.5°, and a power sensor with a sampling frequency of 1Hz.

4. The drainage pipeline inspection and control system based on adaptive path planning according to claim 1, characterized in that, The pipeline navigability dynamic modeling module includes: The internal space of the pipeline is divided into a three-dimensional grid array. Based on the multi-dimensional environmental information dataset, the type of obstacle in each grid is identified according to a preset obstacle determination threshold. The grid passage status is classified and labeled as impassable, restricted passage, free passage, or undetected. For each passable grid cell, the thickness of the sediment, the impact force of the water flow, and the proximity factor of the defect area are extracted, and after normalization, a weighted calculation is performed to obtain the passage cost of the grid cell. Based on the passage status and passage cost of all grids, a visualized pipeline accessibility probability map is generated and dynamically updated according to distance, time, or event triggering conditions.

5. The drainage pipeline inspection and control system based on adaptive path planning according to claim 4, characterized in that, The obstacle determination threshold identification includes: A blockage is defined as follows: a blockage is defined as a sediment thickness ≥ 8 cm; a pipe wall defect depth > 3 mm or a defect length exceeding 50% of the grid side length is defined as a pipe wall defect; a water flow velocity > preset velocity threshold or an angle between the water flow direction and the robot's travel direction > preset angle threshold is defined as a fluid obstacle; and a hydrogen sulfide concentration > 10 ppm is defined as a gas obstacle.

6. The drainage pipeline inspection and control system based on adaptive path planning according to claim 1, characterized in that, The calculation of the passage cost includes: The min-max normalization algorithm is used to map the sediment thickness, water flow impact force, and proximity of the defect area to the [0,1] interval, respectively. The normalization formulas are as follows: Normalized value of sediment thickness = (actual sediment thickness - minimum sediment thickness) / (maximum sediment thickness - minimum sediment thickness); Normalized value of water flow impact force = (actual water flow impact force - minimum water flow impact force) / (maximum water flow impact force - minimum water flow impact force); The normalized value of the proximity of the defect area is calculated as (maximum proximity distance - actual proximity distance) / (maximum proximity distance - minimum proximity distance); the final passage cost is calculated using a weighted linear combination model.

7. The drainage pipeline inspection and control system based on adaptive path planning according to claim 1, characterized in that, The multi-objective adaptive path planning engine module specifically includes: The inspection task instruction is parsed to obtain parameters such as mandatory inspection area, time limit, confidence level limit, energy consumption limit, and coverage requirement, and the multiple optimization objectives are prioritized. Construct a multi-objective optimization function with the objectives of minimizing total inspection time, maximizing defect detection confidence, minimizing robot energy consumption, and maximizing cross-sectional coverage, and set constraints for path planning; Adopting improved A The algorithm performs a global path search on the pipeline traversability probability map, the improved A The algorithm's heuristic function combines the distance between nodes with the cost of passage. Based on the key node sequence of the global path, a model predictive control algorithm is used for local rolling optimization and dynamic adjustment to generate the final inspection path.

8. The drainage pipeline inspection and control system based on adaptive path planning according to claim 1, characterized in that, The closed-loop execution and learning module specifically includes: The inspection path is converted into motion control commands and sent to the inspection robot, and the robot's position, attitude, environmental parameters and task execution parameters are monitored in real time. The deviation between the planned value and the actual monitored value is calculated. When the deviation reaches the warning threshold, local path replanning is performed through the model predictive control algorithm. When the deviation reaches the emergency threshold or a sudden obstacle is detected, global path replanning is triggered. After the inspection task is completed, the validity of the inspection data is verified, and based on the actual execution data, the weight coefficients in the calculation model of the passage cost and the parameters of the optimized path planning algorithm are updated using a fitting method.

9. The drainage pipeline inspection and control system based on adaptive path planning according to claim 3, characterized in that, The installation of the multimodal sensor array includes: The 3D LiDAR and the industrial camera are mounted on the front end of the inspection robot, and the scanning angle and lens orientation are configured to spatially align the image data with the point cloud data; the ultrasonic flow meter is mounted on both sides of the robot to measure the flow velocity at multiple measuring points on the cross-section of the pipe; the multi-gas sensor module is mounted on the top of the robot, with its air inlet facing the direction of the pipe airflow.

Images