Unmanned aerial vehicle pipeline inspection control method for inspection of branched pipelines

Abstract

The present application relates to a kind of unmanned aerial vehicle pipe inspection control methods for bifurcated pipeline inspection, comprising: obtaining the data of various sensors output on unmanned aerial vehicle;By comparing the point cloud data of current time and the point cloud data of previous time obtained by unmanned aerial vehicle, whether unmanned aerial vehicle is in gradual change bend section or abrupt change bend section is judged, and different yaw angle adjustment strategies are used according to the judgment result;The difference between the predicted value of the yaw angle of unmanned aerial vehicle and the actual yaw angle is calculated, and the yaw angle compensation is realized;The point cloud data at the inspection target point is obtained by using the single-line laser radar on unmanned aerial vehicle;The point cloud data is cut in four quadrants;The reference of point cloud model is determined;The point cloud model at the inspection target point is released.The present application solves the technical problems such as pipeline inspection fork, and realizes the underground pipeline branch environment control strategy based on real-time yaw angle adjustment.

Description

Technical Field

[0001] This invention belongs to the field of hydropower station safety monitoring, specifically relating to a drone pipeline inspection control method for branching pipeline inspection. Background Technology

[0002] Branching pipelines are commonly found in underground environments of industrial facilities such as large power plants, used for drainage, purification, and decontamination. These environments typically lack adequate lighting, and due to their underground location, GPS signals are often unavailable.

[0003] Taking underground drainage tunnels as an example, these environments are typically constructed of reinforced concrete. The common structure is a three-dimensional structure formed by stretching a cross-section along a straight line or curve, with a relatively complex internal network and numerous branch tunnels. Over time, due to constant water erosion and the accumulation of silt and debris, the walls will inevitably develop impact pits, cracks, and seepage. In severe cases, this can lead to the collapse of sections of the pipe, affecting normal water discharge. If the main section becomes blocked, a large amount of water may accumulate, potentially damaging the entire pipe section and resulting in significant repair and downtime costs. Therefore, regular inspection of the pipe walls is crucial. Because drainage tunnels have varying water volumes throughout the year—dry during the dry season and large volumes during the flood season—they are prone to erosion, collapse, and accumulation. Using small vehicles or unmanned boats for inspection is unsuitable, inefficient, and risky. Furthermore, the tunnels contain numerous branch tunnels and feature slopes, bends, and vertical curves such as cross or T-shapes. Therefore, when a drone used for inspection encounters a non-unique target point, a strategy is needed to prevent the drone from repeatedly flying in a single direction. Summary of the Invention

[0004] The purpose of this invention is to address the aforementioned problems by providing a UAV pipeline inspection control method for branching pipeline inspection. This method involves equipping the UAV with a single-line lidar, a monocular camera, a binocular camera, and a single-point lidar sensor. Considering the unique three-dimensional structure of pipeline environments such as branches, forks, and bends, a control strategy based on real-time yaw angle adjustment in underground pipeline branch environments ensures optimal selection of the UAV within the branch environment and effective flight through the elimination of duplicate routes. Furthermore, the yaw angle of the UAV is adjusted based on its own angle relative to the pipeline's central axis.

[0005] The technical solution of this invention is a UAV pipeline inspection control method for branching pipeline inspection, comprising the following steps: Step 1: Acquire data output from various sensors on the drone; Step 2: By comparing the point cloud data acquired by the UAV at the current moment with the point cloud data at the previous moment, determine whether the UAV is in a gradual curve or a steep curve, and adopt different yaw angle adjustment strategies according to the judgment results. Step 3: Calculate the difference between the predicted yaw angle and the actual yaw angle of the UAV to achieve yaw angle compensation; Step 4: Use the single-line lidar on the drone to obtain point cloud data at the inspection target point; Step 5: Perform four-quadrant segmentation on the point cloud data obtained in Step 4; Step 6: Determine the validity of the four-quadrant point cloud obtained in Step 5, and determine the baseline of the point cloud model based on the determination result; Step 7: Publish the point cloud model of the inspection target point.

[0006] In step 2, the point cloud data at the current moment is obtained by the forward-facing single-point lidar and the cross-sectional single-line lidar on the UAV. The point cloud data at the previous moment is compared with the point cloud data to determine the scene in which the UAV is currently located. Different yaw angle adjustment strategies are adopted according to whether the scene is a gradual curve or a steep curve.

[0007] Preferably, for a gradually changing curve, a strategy of "adjusting the yaw angle while flying forward" is adopted. For a steep curve, where the space in front of the UAV is small and it cannot continue to move forward, a strategy of "fixed-point rotation" is adopted.

[0008] Preferably, in step 6, the validity of the four-quadrant point cloud obtained in step 5 is determined. If the number of point clouds is small or the number of outliers in the point cloud is large, then the point cloud is determined to be invalid.

[0009] Furthermore, in step 6, the validity of the left and right point clouds is determined respectively, specifically including: If both the left and right point clouds are valid, then calculate the centroid of the entire point cloud and use the centroid of the point cloud as the reference for the point cloud model. If the left point cloud is valid and the right point cloud is invalid, then calculate the mean point of the left point cloud and use the mean point of the left point cloud as the benchmark of the point cloud model. If the left point cloud is invalid but the right point cloud is valid, then calculate the mean point of the right point cloud and use the mean point of the right point cloud as the benchmark of the point cloud model. If both the left and right point clouds fail, the centroid of the previous moment is used as the reference for the point cloud model, and the centroid of the current moment is recorded.

[0010] Preferably, in step 7, a point cloud model selection strategy for repeated target points is adopted, specifically as follows: (1) Record the location of special scenes and determine whether they appear in special branch locations by Euclidean distance, and remove duplicates from the point cloud models of special branch locations; (2) Use computer programs to compare and determine whether point cloud models are duplicates, and delete duplicate point cloud models.

[0011] Compared with the prior art, the beneficial effects of the present invention include: 1) This invention solves technical problems such as branching in pipeline inspection and realizes an environmental control strategy for underground pipeline branches based on real-time yaw angle adjustment.

[0012] 2) The method of this invention ensures real-time flight safety and non-repetitive flight of the UAV in pipeline branch scenarios. Furthermore, it enables real-time adjustment of the UAV's yaw angle. Attached Figure Description

[0013] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0014] Figure 1 This is a schematic diagram illustrating the determination of a pipe bend scenario according to an embodiment of the present invention.

[0015] Figure 2 This is a schematic diagram of the abrupt and gradual transition sections of the pipeline according to an embodiment of the present invention.

[0016] Figure 3 This is a schematic diagram illustrating the determination of the validity of a four-quadrant point cloud and the determination of a point cloud reference in an embodiment of the present invention.

[0017] Figure 4 This is a schematic diagram illustrating the effective point cloud in an embodiment of the present invention.

[0018] Figure 5 This is a schematic diagram of the point cloud failure of a right-angle bend segment according to an embodiment of the present invention.

[0019] Figure 6 This is a schematic diagram illustrating the validity determination of point clouds at T-junctions and crossroads according to an embodiment of the present invention.

[0020] Figure 7 This is a schematic diagram of the yaw angle according to an embodiment of the present invention.

[0021] Figure 8 This is a schematic diagram of yaw angle dilution according to an embodiment of the present invention.

[0022] Figure 9 This is a schematic diagram of yaw angle negative feedback according to an embodiment of the present invention. Figure 10 This is a point cloud diagram of the overall pipeline in an embodiment of the present invention. Detailed Implementation

[0023] The implementation plan is mainly divided into the following three parts: the judgment of gradual curves and steep curves and the adjustment of the UAV's real-time yaw angle; the target point release strategy in environments such as side roads and crossroads; and the optimal deduplication selection in the case of multiple repeated target points.

[0024] The difference between gradual and abrupt curves lies in their direct impact on the real-time adjustment strategy for the yaw angle—that is, the corresponding servo rotation and stationary rotation, respectively. The scenarios faced by the drone include... Figure 2 As shown, this indicates whether there is a sudden change in the angle of the central axis of the pipe. Figure 2 The left and middle pipes shown both exhibit abrupt changes in their central axis, with right-angle bends being more common; the right pipe, on the other hand, shows a gradual change.

[0025] The drone, equipped with a single-line lidar, binocular camera, and single-point lidar, can make judgments on multiple scenes and adjust its pose in real time under the judgment conditions.

[0026] The UAV pipeline inspection control method for branching pipeline inspection includes: Step 1: Acquire data output from various sensors on the drone; Step 2: By comparing the point cloud data acquired by the UAV at the current moment with the point cloud data at the previous moment, determine whether the UAV is in a gradual curve or a steep curve, and adopt different yaw angle adjustment strategies according to the judgment results. Step 3: Calculate the difference between the predicted yaw angle and the actual yaw angle of the UAV to achieve yaw angle compensation; Step 4: Use the single-line lidar on the drone to obtain point cloud data at the inspection target point; Step 5: Perform four-quadrant segmentation on the point cloud data obtained in Step 4; Step 6: Determine the validity of the four-quadrant point cloud obtained in Step 5, and determine the baseline of the point cloud model based on the determination result; Step 7: Publish the point cloud model of the inspection target point.

[0027] Based on real-time data from the UAV's forward-facing single-point laser and single-line lidar, as well as various data from previous moments or more, scene judgment is made by comparing the data from the previous moment with the current moment. The overall process logic is as follows: Figure 1 As shown.

[0028] The real-time yaw angle adjustment of the drone mainly relies on the comparison between the current pose output of the binocular camera and the current centerline calculated angle, with the goal of ensuring that the drone's nose is always along the centerline of the pipeline.

[0029] After implementing real-time yaw angle adjustment, the next step is to implement the path environment judgment logic for the UAV, i.e., optimal selection. Taking a left-right path as an example, the discussion of the vertical and horizontal path environments for the UAV is similar.

[0030] Point cloud data of the inspection target points is obtained using a single-line lidar on a drone. The point cloud data is then segmented into four quadrants to determine the validity of the four-quadrant point cloud. Based on the determination results, the baseline of the point cloud model is established, such as... Figure 3 As shown.

[0031] If both the left and right point clouds are valid, calculate the centroid of the entire point cloud and use it as the reference for the point cloud model. If the left point cloud is valid but the right point cloud is invalid, calculate the mean point of the left point cloud and use it as the reference for the point cloud model. If the left point cloud is invalid but the right point cloud is valid, calculate the mean point of the right point cloud and use it as the reference for the point cloud model. If both the left and right point clouds are invalid, use the centroid of the previous moment as the reference for the point cloud model and record the centroid of the current moment.

[0032] In the target point deployment process, the drone's position along the axis is determined by adding a step size to the current position, and does not depend on the input of the four-quadrant point cloud. The length of the step size depends on the required drone flight speed. Using a certain reference point in the flowchart refers to the fact that in two directions other than the axis, the drone needs to use the point cloud data in that direction as a reference to ensure that the drone is in the center of the cave environment. The typically chosen value is the calculated value of the point cloud data on one side plus half the cave width.

[0033] The next problem to solve is the selection strategy for drones when facing repeated target points in the same environment. For example, when a drone appears at an intersection, excluding the rear, it can choose to fly in three directions: forward, left, and right. In this case, if the drone simply adopts a target point priority strategy, such as forward having the highest priority, followed by the left, and the right the last, the drone will always fly ahead when passing through the intersection, causing the pipes on both sides to be missed during inspection.

[0034] Therefore, its overall inspection deduplication logic is as follows: (1) Setting the priority of drone branch scenarios, such as the priority of front, left and right; (2) Record the location of special scenes; (3) Determine whether it appears in a special branch location by using Euclidean distance; (4) Code to remove duplicate target points; (5) Issue inspection notices for target points.

[0035] In the embodiment, the drone can autonomously travel along the axis in various forks and other scenarios to achieve the inspection target. The failure of point cloud refers to the lidar measurement distance being within a certain range, or in such a corridor, when the distance between the buildings on both sides of the drone exceeds a certain distance, such as the effective width of the corridor, it is considered a failure value.

[0036] The following scenarios are used to determine whether a branching region has been entered by analyzing the failure of the point cloud, and to make the branching decision-making plan mentioned in the patent: 1) For straight or gradually changing corridors, point clouds on both sides do not fail, such as... Figure 4 As shown.

[0037] 2) In scenarios involving right-angle turns, if, upon entering the vicinity of the intersection, the distance between the point clouds on one side is significantly greater than that on the other side, then the side with the greater distance is considered invalid. Figure 5 As shown.

[0038] 3) When entering T-junctions and crossroads, both sides of the drone are considered ineffective. When the drone enters a T-junction or crossroads, if the distance from both sides of the drone to the point cloud is large, exceeding a certain threshold such as 3-4 times the width of the corridor, then the point clouds on both sides of the drone are considered ineffective. Figure 6 As shown.

[0039] By using a single-line lidar on a drone, point cloud information of the drone on a cross section can be obtained in real time during actual flight. Assuming the existence of an ordered set of centroids, and considering that only the yaw angle needs to be calculated, we can consider only the electric clouds on the left and right sides of the single-line lidar, or after calculating the centroids, only the coordinate values ​​on both sides can be considered.

[0040] Taking the x-direction as the forward direction of the drone and the y-direction as the left and right direction as an example, the yaw angle of the drone is shown in Figure 7, where P is the forward direction of the drone. i (x i ,y i ,z i ) and P i-1 (x i-1 ,y i-1 ,z i-1 ) represents the coordinates of two adjacent positions of the UAV in the world coordinate system.

[0041] Yaw angle: (1)

[0042] In the formula: The yaw angle is the angle between the yaw angle and the x-axis, and w represents the yaw rate.

[0043] According to equation (1), considering that spatial three-dimensional structures are usually continuous closed structures and culverts and pipes rarely have abrupt changes, the yaw angle at the next moment can be predicted based on the centroid information of the first n frames during flight, that is, the yaw angle at the n+1th frame.

[0044] For point set P, then:

[0045] The first calculation in equation (1) transforms the set of centroid points into a set of yaw angles. :

[0046]

[0047] Based on mean estimation, when the weights of the first n points are consistent, the yaw angle at the nth point is... It can be represented as:

[0048] (2)

[0049] When the yaw angle changes, based on experience and practical analysis, the weight of the yaw angle at the instant of change should be greater, and the weight of the yaw angle closer to the current moment should also be greater. Therefore, setting the weight of the first n centroids to be the same will result in the first n-1 points diluting the yaw angle of the nth point. Figure 8 As shown.

[0050] When the drone is at position A, its subsequent yaw angle should be the yaw angle at position A. However, if an equal-weighted estimation method is used, the yaw angle will be weakened to... .

[0051] Furthermore, the larger the K value, that is, the more centroid data are taken into consideration, the greater the deviation between the estimated value and the true value will be.

[0052] To address the problems caused by equal weights, the equal weight approach is optimized by rewriting equation (2) as follows:

[0053] Wherein, the weighting coefficients satisfy the following two conditions:

[0054] In actual flight, a difference inevitably arises between the predicted and actual values. Assuming... This represents the expected value. If the actual value is represented, then the difference between the two can be expressed as:

[0055]

[0056] To make the predicted values ​​closer to the actual values, As negative feedback added to the entire yaw angle prediction system, such as Figure 9 As shown.

[0057] The overall prediction expression can then be written as:

[0058] The point cloud model of the pipeline obtained in the embodiment is as follows: Figure 10 As shown.

Claims

1. A UAV pipeline inspection control method for inspecting branching pipelines, characterized in that, The drone used for inspecting branching pipelines is equipped with a single-line lidar, a monocular camera, a binocular camera, and a single-point lidar sensor. The drone pipeline inspection control method includes the following steps: Step 1: Acquire data output from various sensors on the drone; Step 2: By comparing the point cloud data acquired by the UAV at the current moment with the point cloud data at the previous moment, determine whether the UAV is in a gradual curve or a steep curve, and adopt different yaw angle adjustment strategies according to the judgment results. Step 3: Calculate the difference between the predicted yaw angle and the actual yaw angle of the UAV to achieve yaw angle compensation; Step 4: Use the single-line lidar on the drone to obtain point cloud data at the inspection target point; Step 5: Perform four-quadrant segmentation on the point cloud data obtained in Step 4; Step 6: Determine the validity of the four-quadrant point cloud obtained in Step 5, and determine the baseline of the point cloud model based on the determination result; Step 7: Publish the point cloud model of the inspection target point; In step 6, the validity of the four-quadrant point cloud obtained in step 5 is determined. If the number of point clouds is small or the number of outliers in the point cloud is large, the point cloud is determined to be invalid. In step 6, the validity of the left and right point clouds is determined respectively, specifically including: If both the left and right point clouds are valid, then calculate the centroid of the entire point cloud and use the centroid of the point cloud as the reference for the point cloud model. If the left point cloud is valid and the right point cloud is invalid, then calculate the mean point of the left point cloud and use the mean point of the left point cloud as the benchmark of the point cloud model. If the left point cloud is invalid but the right point cloud is valid, then calculate the mean point of the right point cloud and use the mean point of the right point cloud as the benchmark of the point cloud model. If both the left and right point clouds fail, the centroid of the previous moment is used as the reference for the point cloud model, and the centroid of the current moment is recorded.

2. The UAV pipeline inspection control method for branching pipeline inspection according to claim 1, characterized in that, In step 2, the point cloud data at the current moment is obtained by the forward single-point lidar and cross-sectional single-line lidar on the UAV, and compared with the point cloud data at the previous moment to determine the scene in which the UAV is currently located. Different yaw angle adjustment strategies are adopted according to whether the scene is a gradual curve or a steep curve. If it is a gradual curve, the strategy of "adjusting the yaw angle while flying forward" is adopted. If it is a steep curve, the space in front of the drone is small, so it cannot continue to move forward and adopts a "fixed-point rotation" strategy.

3. The UAV pipeline inspection control method for branching pipeline inspection according to claim 1 or 2, characterized in that, In step 7, a point cloud model selection strategy is adopted for repeated target points, specifically as follows: (1) Record the location of special scenes and determine whether they appear at the intersection by Euclidean distance, and remove duplicates from the point cloud model of the intersection; (2) Use computer programs to compare and determine whether point cloud models are duplicates, and delete duplicate point cloud models.

Images