Unmanned aerial vehicle flight control method and apparatus

Abstract

The present disclosure provides a kind of unmanned aerial vehicle flight control method and device, it is related to unmanned aerial vehicle autonomous flight control technical field.The method includes the topological features of the skeleton curve family based on the structure of pipe row surface is constructed, and the surface grid coordinate system containing along the skeleton direction parameter and normal bias parameter is established;The step distance required for pipe row surface coverage and the constraint condition of unmanned aerial vehicle attitude control are determined in combination with the performance parameters of millimeter wave radar on unmanned aerial vehicle and the preset coverage requirement;Based on step distance, generate coverage strip in surface grid coordinate system, and fit coverage strip as smooth trajectory, wherein, smooth trajectory meets the requirement of constraint condition and minimum turning radius;Control unmanned aerial vehicle to fly along smooth trajectory, and carry out inspection to pipe row structure in the process of flight, output detection result.The present disclosure can improve the autonomous coverage flight capability and navigation positioning precision of unmanned aerial vehicle in complex surface structure environment, avoid appearing missed detection.

Description

Technical Field

[0001] This disclosure relates to the field of autonomous flight control technology for unmanned aerial vehicles (UAVs), and in particular to a UAV flight control method and apparatus. Background Technology

[0002] In industrial inspection scenarios, flight control systems (FCS) and path planning technologies have begun to be applied to inspection tasks in complex curved structural spaces. A typical example is the use of drones for inspection in confined environments such as boiler water-cooled walls. By equipping drones with millimeter-wave radar or high-definition cameras, autonomous navigation and data collection can be achieved in narrow spaces such as inside furnaces, reducing reliance on close-range manual inspection. However, a common issue with these technologies is that drones are prone to missing certain areas during inspections. Summary of the Invention

[0003] To overcome the problems existing in related technologies, this disclosure provides a method and apparatus for controlling the flight of an unmanned aerial vehicle (UAV).

[0004] According to a first aspect of the present disclosure, a drone flight control method is provided, applied to pipeline inspection, the method comprising:

[0005] Based on the topological features of the pipe-pile surface structure, a family of skeleton curves is constructed, and a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters is established.

[0006] Based on the performance parameters of the millimeter-wave radar on the UAV and the preset coverage requirements, the step distance required for the pipe array curved surface coverage and the constraints of the UAV attitude control are determined.

[0007] Based on the step size, a covering strip is generated in the surface grid coordinate system, and the covering strip is fitted into a smooth trajectory, wherein the smooth trajectory satisfies the constraint conditions and the minimum turning radius requirements;

[0008] Control the drone to fly along a smooth trajectory and inspect the pipe structure during flight, outputting the inspection results.

[0009] In some embodiments, the method further includes:

[0010] Calculate the loop closure confidence based on the likelihood between the current scan data and historical scan data of the UAV;

[0011] When the loop closure confidence is less than a preset threshold, the drone is controlled to repeatedly cruise in the current area in order to scan the scan data from different perspectives in the current area, and the loop closure confidence is recalculated in combination with the scan data from different perspectives to make a loop closure confidence judgment.

[0012] When the loop closure confidence is greater than or equal to a preset threshold, global attitude graph optimization is performed.

[0013] In some embodiments, the construction of a family of skeleton curves based on the topological features of the pipe-pile surface structure, and the establishment of a surface mesh coordinate system including parameters along the skeleton direction and normal offset parameters, includes:

[0014] Extract the central axis of the curved surface structure and generate a family of skeleton curves using spline fitting;

[0015] Using the equal arc length parameter of the skeleton curve as the parameter along the skeleton direction, and the distance perpendicular to the skeleton curve as the normal offset parameter, a local surface patch is constructed.

[0016] Apply continuous splicing conditions to adjacent local surface patches to form an overall surface grid coordinate system.

[0017] In some embodiments, determining the step size required for pipe array surface coverage and the constraints for UAV attitude control by combining the performance parameters of the millimeter-wave radar on the UAV with preset coverage requirements includes:

[0018] Based on the geometric relationship between radar pose and curved surface, the half-power width of geometric radar is used to determine the projection width of radar beam on curved surface.

[0019] Based on the preset minimum coverage overlap and the projection width, determine the coverage step size perpendicular to the skeleton direction;

[0020] Based on the maximum allowable phase window, radar wavelength, and equivalent lever arm, the phase stability constraint is transformed into the aforementioned constraint condition.

[0021] In some embodiments, generating a cover strip within the surface mesh coordinate system based on the step size and fitting the cover strip to a smooth trajectory includes:

[0022] Within the curved mesh coordinate system, parallel coverage strips are generated with a coverage step perpendicular to the skeleton direction; adjacent coverage strips are connected using a serpentine path strategy, and redundant overlapping areas are set at the ends of the coverage strips.

[0023] Each coverage strip is fitted as a piecewise curve;

[0024] With the goal of suppressing curve curvature and acceleration, and in combination with the constraints and minimum turning radius requirements, each segmented curve is smoothed and optimized to generate a smooth trajectory.

[0025] In some embodiments, fitting each coverage strip to a piecewise curve includes:

[0026] Each coverage strip is fitted with a piecewise cubic Bézier curve using the following formula:

[0027] ;

[0028] in, For Bézier curves, It is the standardized arc length parameter along the curve. These are all control points in a cubic Bézier curve.

[0029] In some embodiments, the step of optimizing the smoothness of each segmented curve to generate a smooth trajectory, with the goal of suppressing curve curvature and jerk, in conjunction with the constraints and minimum turning radius requirements, includes:

[0030] Define the smoothing cost function:

[0031] ;

[0032] in, For smoothing cost function, This is the curvature weighting coefficient. For jerk weighting coefficient, To accelerate, For curvature, , for The first derivative, for The second derivative;

[0033] By adjusting the position of the control points, the smoothing cost function is minimized, and the generated smooth curve satisfies the constraints and minimum turning radius requirements, thereby achieving smooth optimization of the piecewise curve.

[0034] In some embodiments, the method further includes:

[0035] Define a joint cost function for coverage gap, overlap deviation, wall distance error, phase error, and energy consumption:

[0036] ;

[0037] in, For the joint cost function, For parameter field path, Used to measure the difference between actual overlap and target overlap. Areas not covered by penalties This is the wall distance error. This is a phase out-of-bounds term. For phase window, For the maximum allowed phase window, To control the input, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item;

[0038] During the flight of the UAV, the flight parameters of the UAV are optimized in a finite time domain based on the model predictive control algorithm to minimize the joint cost function and satisfy the preset constraints.

[0039] In some embodiments, after constructing a family of skeleton curves based on the topological features of the pipe-row surface structure and establishing a surface mesh coordinate system including parameters along the skeleton direction and normal offset parameters, the method further includes:

[0040] Identify openings, corners, and attached component areas in the curved structure and mark them as impassable areas in the curved grid coordinate system.

[0041] According to a second aspect of this disclosure, a drone flight control device is provided for pipe rack inspection, the device comprising:

[0042] A module is established to construct a family of skeleton curves based on the topological features of the pipe-pile surface structure, and to establish a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters.

[0043] The determination module is used to combine the performance parameters of the millimeter-wave radar on the UAV with the preset coverage requirements to determine the step distance required for the coverage of the pipe array curved surface and the constraints of the UAV attitude control.

[0044] The fitting module is used to generate a covering strip in the surface grid coordinate system based on the step size, and fit the covering strip into a smooth trajectory, wherein the smooth trajectory satisfies the constraint conditions and the minimum turning radius requirements;

[0045] The control module is used to control the UAV to fly along a smooth trajectory and to inspect the pipe structure during flight, outputting the inspection results.

[0046] The technical solutions provided in this disclosure can include the following beneficial effects:

[0047] This disclosure allows for the construction of a family of skeleton curves based on the topological features of the pipe array curved surface structure, and the establishment of a curved surface grid coordinate system containing parameters along the skeleton direction and normal offset parameters. Combining the performance parameters of the millimeter-wave radar on the UAV with preset coverage requirements, the required step distance for pipe array curved surface coverage and the constraints for UAV attitude control are determined. Based on the step distance, coverage strips are generated within the curved surface grid coordinate system, and these strips are fitted into a smooth trajectory, wherein the smooth trajectory satisfies the constraints and minimum turning radius requirements. The UAV is controlled to fly along the smooth trajectory and inspects the pipe array structure during flight, outputting the inspection results. This disclosure can improve the autonomous coverage flight capability and navigation and positioning accuracy of UAVs in complex curved surface structure environments, avoiding missed detections.

[0048] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0049] Figure 1 The diagram shows a flowchart of a drone flight control method according to an embodiment of this disclosure.

[0050] Figure 2 This diagram illustrates the construction of a surface coordinate system according to an embodiment of the present disclosure.

[0051] Figure 3 A schematic diagram of the state machine of a UAV scanning planning generation method in an embodiment of this disclosure is shown.

[0052] Figure 4 A schematic diagram of the state machine of a loop closure detection method according to an embodiment of this disclosure is shown.

[0053] Figure 5 A schematic diagram of the overall architecture of an embodiment of this disclosure is shown.

[0054] Figure 6 A schematic diagram of the structure of a drone flight control device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0055] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0056] In industrial inspection scenarios, flight control systems (FCS) and path planning technologies have begun to be applied to inspection tasks in complex curved structural spaces. A typical example is the use of drones for inspection in confined environments such as boiler water-cooled walls. By equipping drones with millimeter-wave radar or high-definition cameras, autonomous navigation and data collection can be achieved in narrow spaces such as inside furnaces, reducing reliance on close-range manual inspection. Currently, a common approach is to plan scanning paths that cover the entire wall surface, such as performing "S"-shaped reciprocating flights along the water-cooled wall surface, to ensure comprehensive coverage of the curved area. Adjacent coverage bands along the drone's flight path should maintain a certain degree of overlap to prevent blind spots and missed inspections.

[0057] However, traditional navigation systems often generate paths centered on discrete target points, neglecting the specific flight quality requirements of millimeter-wave imaging missions. Path planning often focuses only on the geometrically shortest route from one point to another, without fully considering key constraints such as imaging overlap ratio, phase stability, and attitude stability. The lack of these quality constraints can easily lead to distortion in UAV scan images, incomplete coverage of the target area, and ultimately, a decrease in detection accuracy.

[0058] In a three-dimensional environment, path planning and flight control coordination face numerous challenges. The normal distance to curved surfaces varies drastically, and structures such as water-cooled wall tube arrays necessitate frequent adjustments to the distance between the UAV and the surface. Complexly curved surfaces require the flight path to closely follow the contour; a lack of smooth path optimization can lead to sudden changes in aircraft attitude (attitude abrupt changes). Even when using methods such as Bezier curves to generate smooth flight paths, abrupt attitude adjustments are unavoidable during real-time flight following curved surfaces, thus testing the responsiveness of the flight control system.

[0059] Furthermore, navigation in enclosed curved environments is also constrained by various sensor signal interference factors. Enclosed spaces such as boiler furnaces lack GPS signals and have low light levels, making conventional satellite and visual navigation ineffective. Metal walls and diffused dust introduce significant electromagnetic and optical noise, leading to decreased positioning and obstacle avoidance accuracy. These environmental factors further exacerbate the difficulty of flight control in complex curved spaces, making it difficult for the system to maintain attitude and navigation stability.

[0060] Furthermore, current loop closure detection and map correction methods are mostly based on global confidence propagation strategies, uniformly propagating and adjusting errors across the entire map. These global methods neglect the robustness of local regions within curved surfaces, fail to perform probabilistic modeling for local repetitive features, and lack optimization for loop closure triggering mechanisms. Therefore, once a loop closure is misidentified, it easily leads to incorrect matching, resulting in a significant increase in accumulated map errors.

[0061] To address the aforementioned challenges, this disclosure focuses on constructing a curved mesh coordinate system based on pipe-pile topology. It further incorporates optimizations in various aspects, such as coverage-optimized path planning, probabilistic modeling mechanisms for loop closure detection, and attitude and trajectory collaborative control, to comprehensively improve the flight control stability and navigation robustness of the system in high-precision imaging and detection tasks in curved space.

[0062] Specifically, this disclosure departs from existing paradigms and proposes a curved mesh coverage and loop closure enhancement flight control system based on pipe-pile topology. The system is dominated by the centerline network of water-cooled wall pipes, constructing a curved mesh coordinate system composed of pipe and normal axes. The "step size and bandwidth" are jointly determined by the millimeter-wave beam model and target resolution requirements. Three quality indicators—overlap, phase window, and wall distance—are explicitly written into the constraint set of flight control and path generation, forming a "mission quality-driven" control closed loop. The core mechanism includes five stages: curved surface modeling, coverage optimization, curve smoothing, loop closure probability and re-cruise triggering, and pose and scan coordination, which are executed cyclically.

[0063] The technical problem this disclosure aims to solve is three types of coupling mismatch. First, the mismatch between path space and physical structure leads to coverage gaps in the pipe array topology. Second, the mismatch between imaging parameters and control variables causes phase instability and SNR degradation due to attitude changes. Third, the mismatch between mapping correction and triggering logic results in a global, one-time correction of amplified errors. This disclosure aligns path generation with the physical topology using curved mesh coordinates, aligns control objectives with imaging requirements using quality constraints, and aligns the triggering mechanism with local robustness using a loop closure probability map, thereby uniformly solving these three types of mismatch.

[0064] The core solution disclosed herein comprises three main components. First, curved mesh modeling and coverage optimization. A two-parameter curved mesh is established using the pipe centerline as the dominant curvature direction and the normal distance as the wall distance state. Based on the millimeter-wave radar beamwidth, sidelobe suppression, and desired spatial resolution, the minimum lateral step size and inter-band overlap threshold are calculated to generate coverage bands with "equal arc length and equal overlap." Second, curve smoothing and attitude coordination. Long flight paths are smoothed on the curved mesh using Bezier curves, constraining the minimum turning radius and attitude angular velocity. Combined with gimbal pointing and airframe angular velocity coordination, phase broadening is suppressed. Third, loop closure probability enhancement and re-cruise triggering. Instead of immediate global correction, loop closure confidence is established locally and updated using Bayesian methods. When confidence is insufficient, re-cruise is triggered, correcting only the local submap. When confidence crosses the threshold, the trajectory and map are uniformly optimized to reduce mismatch risk.

[0065] The following will describe the exemplary implementation method in detail with reference to the accompanying drawings and embodiments.

[0066] Figure 1 This diagram illustrates a flow chart of a UAV flight control method according to an embodiment of the present disclosure, applied to pipeline inspection. Figure 1 As shown in the embodiments of this disclosure, the UAV flight control method includes the following steps.

[0067] S110, based on the topological features of the pipe-pile surface structure, constructs a family of skeleton curves and establishes a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters.

[0068] S120, combining the performance parameters of the millimeter-wave radar on the UAV with the preset coverage requirements, determines the step distance required for the curved surface coverage of the pipe array and the constraints for the attitude control of the UAV.

[0069] S130, based on the step size, generates a covering strip in the surface grid coordinate system and fits the covering strip into a smooth trajectory, wherein the smooth trajectory satisfies the constraints and the minimum turning radius requirements.

[0070] The S140 controls the drone to fly along a smooth trajectory and inspects the pipe structure during flight, outputting the inspection results.

[0071] In some embodiments, S101 can be implemented in the following manner:

[0072] S111: Extract the central axis of the curved surface structure and generate a family of skeleton curves using spline fitting.

[0073] S112 uses the equal arc length parameter of the skeleton curve as the parameter along the skeleton direction and the distance perpendicular to the skeleton curve as the normal offset parameter to construct a local surface patch.

[0074] S113 applies a continuous splicing condition to adjacent local surface patches to form an overall surface grid coordinate system.

[0075] For example, after constructing the surface mesh coordinate system, it is also possible to identify the openings, corners and attached component areas in the surface structure and mark them as impassable areas in the surface mesh coordinate system.

[0076] Specifically, to align the path space of the UAV with the physical structure of the tube bank, this disclosure constructs the water-cooled wall tube bank geometry as follows: Figure 2 The surface parameter domain shown .set up The centerline of the pipe along the dominant direction of the pipe bank is obtained by spline fitting of the field point cloud and the design prior. This indicates that the pipe centerline direction is offset. By reparameterizing with equal arc lengths, the parameter step size is made approximately consistent with the physical arc length. Let... for If the unit normal is given, then the surface in the neighborhood of the centerline can be represented as:

[0077] ;

[0078] in, This is the normal bias. The corresponding basis vectors are... , The first basic form is:

[0079] ;

[0080] in, The coefficient represents the square of the length of an infinitesimal line element on the surface. Indicates parameters The magnitude of the contribution of the change in direction to the actual change in the position of the curved surface. ;coefficient Indicates parameters and Degree of coupling between directional changes ;coefficient Indicates parameters The magnitude of the contribution of the change in direction to the actual change in the position of the curved surface. The three coefficients were obtained by taking the derivative of the fitted curve. Indicates parameters Length in direction, Indicates parameters The length of the direction. Because... Equal arc length reparameterization It is more stable in numerical terms, which helps in the subsequent generation of equal arc length strips.

[0081] Wall distance Defined as:

[0082] ;

[0083] therefore, Its derivative It can be directly used as an inner-loop observation for wall-mounted control. Radius of curvature Depend on The first and second derivatives are calculated to set the minimum turning radius and curvature constraints, where... For curvature; when When reducing, the corresponding area needs to tighten the upper limit of turning and acceleration at the planning layer.

[0084] Considering that a real water-cooled wall is a topology network with multiple pipe centerlines, let's denote the family of skeleton curves. ,in, Indicates the first Centerline of the strip (skeleton curve). This represents the total number of skeleton curves. In its engineering implementation, this disclosure employs piecewise parameterization: for each pipe centerline... Constructing local surface patches Adjacent patches are joined at the boundary with continuity in both normal and tangential directions as the splicing condition, and an infeasibility domain mask is applied at the parameter domain boundary to handle openings, attached components, and corners. The resulting Surface Mesh Coordinate System preserves the pipe row topology and provides a stable metric for the subsequent generation of cover strips with "equal arc length and equal overlap".

[0085] To ensure that coverage metrics and imaging constraints are computable within the parameter domain, this disclosure... Construct several equal offset curves in the direction (equal to the direction) (line), and draw normal segments at sampling points of equal arc length as mesh. Wire; Equal arc length sampling points in the direction This grid is used to discretize the start and join positions of the stripes. Subsequent steps will use this grid as a basis to derive the projected width of the millimeter-wave beam on the curved surface. Step distance Co-constraints are applied with the attitude-phase window, thereby generating overlapping coverage strips and smooth tracks.

[0086] In some embodiments, S120 can be implemented in the following manner:

[0087] S121, based on the geometric relationship between the radar pose and the curved surface, and the half-power width of the geometric radar, the projection width of the radar beam on the curved surface is determined.

[0088] S122, determine the coverage step size perpendicular to the skeleton direction based on the preset minimum coverage overlap and projection width.

[0089] S123 transforms the phase stability constraint into a constraint condition based on the maximum allowable phase window, radar wavelength, and equivalent lever arm.

[0090] Understandably, the purpose of steps S121 to S123 is to establish the UAV's scan plan (e.g., coverage step size and various constraints) in the curved mesh coordinate system. For details, please refer to... Figure 3 The above steps can be achieved in the following way:

[0091] First, geometric ingestion and incidence estimation are performed on the surface for each... Calculate the normal clearance With the angle of incidence This provides geometric quantities for subsequent footprint projection. Since the boiler water-cooled wall is a curved metal surface with varying curvature, minute attitude disturbances of the carrier and gimbal can cause rapid changes in the incident angle. This module, based on the established surface parameterization... Above, in the pose of millimeter-wave radar With visual axis Projected onto the local normal system, a unified approach is given. and Because the calculations are performed directly within the surface coordinate system, the geometric quantities can be consistent with the subsequent coverage measurements, reducing the deviations caused by planar approximations.

[0092] ;

[0093] in, for and Encapsulated vectors for The wall point, For unit normal direction, The beam line-of-sight unit vector, This is the normal distance.

[0094] Next, antenna and beam parameterization can be performed, that is, characterizing the main lobe pattern of the millimeter-wave radar with a small number of parameters, which facilitates analytical calculation and conservative estimation. Since the actual pattern is affected by frequency variations and sidelobes, directly using a full-spectrum model is not conducive to online calculation. A half-power beam width (HPBW) model can be used instead. The core of this approach is an equivalent main lobe model, with side lobe levels (SLL) incorporated into subsequent thresholds as a safety factor. This method ensures parameter stability and ease of calibration, and naturally aligns with task overlap and resolution constraints.

[0095] Subsequently, the effective projection width of the main lobe in the normal direction of the surface can be calculated. The footprint varies due to the coupling between the incident angle and the wall distance. Nonlinear change, this method is based on The given approximate analytical solution makes the equations differentiable and computationally inexpensive, facilitating real-time step-by-step solution:

[0096] ;

[0097] Obtain the effective projection width Then, the target overlap and spatial resolution can be mapped to the normal step distance using the following formula. This refers to the coverage step size perpendicular to the skeleton direction. Because high curvature and grazing incidence regions amplify footprint variations, a fixed step size will lead to overlap drift and missed detections. This method uses the minimum coverage overlap... The analytical step size is given, and adaptive tightening is allowed in the high curvature region. Therefore, the task indicators are directly written into the geometric pitch, which has the advantages of stable coverage and reproducibility.

[0098] ;

[0099] At the same time, the maximum allowed phase window (Phase Window) will be... This is converted into an upper limit for attitude variation, which restricts the scanning rhythm. Since attitude perturbations can cause significant changes in line-of-sight and affect phase stability in a near-field metallic cavity, the method employs an equivalent lever arm. The attitude angle change is mapped to a line-of-sight perturbation, and expressed in terms of wavelength. This restricts it to within a phase window. This method can endogenize the imaging quality boundary into control and scheduling, ensuring that "geometric feasibility" and "imaging usability" are achieved simultaneously.

[0100] ;

[0101] in, This represents the change in the UAV's attitude angle between two scans by the millimeter-wave radar. This refers to the wavelength of millimeter-wave radar.

[0102] Therefore, by summarizing the above steps, a step size including the normal direction can be generated. Sampling intervals along the route Scanning speed With gimbal angular velocity The scan plan serves as the direct input for coverage path generation. Specifically, the normal step size... Once determined, the sampling interval along the route Scanning speed and gimbal angular velocity The parameters can be determined collaboratively using methods found in related technologies and in conjunction with those of millimeter-wave radar, and should also satisfy the attitude angle constraints provided by the phase window. .

[0103] This method is based on The normal step distance (crossband pitch) is determined by the phase window and integration time. By jointly limiting the sampling and scanning speed along the path, and employing a slow-moving, high-overlap strategy locally to achieve imaging stability, this method can achieve integrated output of step size and rate, and seamless integration of interface and coverage planning, facilitating engineering implementation and verification.

[0104] In some embodiments, S130 can be implemented in the following manner:

[0105] S131, in the curved mesh coordinate system, generate parallel cover strips according to the cover step size perpendicular to the skeleton direction.

[0106] The adjacent coverage strips are connected using a serpentine path strategy, and the ends of the coverage strips are provided with redundant overlapping areas.

[0107] Specifically, this can be done within the surface mesh coordinate system, in the parameter domain. First, fix Parallel strips are generated and then trimmed based on boundary and obstacle masks. A serpentine splicing strategy is used between strips to reduce the number of turns, while transition strips are inserted at points of abrupt curvature changes to generate cover strips. This represents the mapping of the entire surface region to be scanned in parameter space. For pipe direction (arc length) parameters, The start and end parameters of the scan along the pipe direction are defined. For normal parameters, The start and end parameters of the scan in the pipe normal direction are defined, which is the overall width range of the scan band. This is the index number of the scanned strip. To prevent missed detections between strips at the splicing points, redundant overlapping segments are added to the end segments of adjacent strips. The length of the redundancy depends on the speed-attitude response time of the aircraft at its minimum turning radius.

[0108] S132, fit each coverage strip as a piecewise curve.

[0109] Specifically, each covering strip can be fitted as a piecewise cubic Bézier curve using the following formula:

[0110] ;

[0111] in, For Bézier curves, It is the standardized arc length parameter along the curve. These are all control points in a cubic Bézier curve.

[0112] S133 aims to suppress curve curvature and jerk, and combines constraints and minimum turning radius requirements to perform smooth optimization on each segmented curve to generate a smooth trajectory.

[0113] Specifically, a smoothing cost function can be defined:

[0114] ;

[0115] in, For smoothing cost function, This is the curvature weighting coefficient. For jerk weighting coefficient, To accelerate, For curvature, , for The first derivative, for The second derivative of .

[0116] By adjusting the position of the control points, the smoothing cost function is minimized, and the generated smooth curve satisfies the constraints and minimum turning radius requirements, thus achieving smooth optimization of the piecewise curve.

[0117] For example, attitude-phase coordination can provide inequality constraints based on the upper limits of angular velocity and phase error, and synchronize the gimbal pointing with the body angular velocity to ensure that attitude changes within the phase window do not exceed the limits. Finally, the control point correction is obtained by approximation using quadratic programming or sequential convex optimization, and a smooth trajectory and desired attitude profile are output.

[0118] In some embodiments, the method provided in this disclosure may further include the following steps:

[0119] Define a joint cost function for coverage gap, overlap deviation, wall distance error, phase error, and energy consumption:

[0120] ;

[0121] in, For the joint cost function, For parameter field path, Used to measure the difference between actual overlap and target overlap. Areas not covered by penalties This is the wall distance error. This is the actual wall distance. For imaging wall distance, This is a phase out-of-bounds term. For phase window, For the maximum allowed phase window, To control the input, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item;

[0122] During the flight of the UAV, the flight parameters of the UAV are optimized in a finite time domain using the Model Predictive Control (MPC) algorithm to minimize the joint cost function and satisfy the preset constraints.

[0123] The constraints can be expressed as follows:

[0124] st ;

[0125] ;

[0126] in, This is a dynamic differential equation model of the UAV's motion, describing the UAV's state (position). ,speed angular velocity How to control input It evolved under the influence of [the environment / the environment].

[0127] This limits the drone's attitude angular velocity. upper limit This is to prevent phase instability caused by excessively rapid steering.

[0128] This limits the flight speed of drones. upper limit .

[0129] This limits the range of wall distances, that is, it restricts the wall distance range. Minimum wall distance and maximum wall distance between.

[0130] This limits the minimum turning radius, which is the radius of curvature of the drone's flight path. It must be greater than or equal to the minimum turning radius that the drone can achieve. Otherwise, the path cannot be flown.

[0131] This limits the maximum curvature, which, in conjunction with the minimum turning radius, determines the curvature of the drone's flight path. Not exceeding the maximum curvature .

[0132] In terms of implementation, model predictive control (MPC) can be used for rolling optimization in the finite time domain. The state vector includes position, velocity, attitude and wall distance observations, and the cost weights can be scheduled in real time according to the imaging quality.

[0133] In some embodiments, the method provided in this disclosure may further include the following steps:

[0134] Calculate the loop closure confidence based on the likelihood between the current scan data and historical scan data of the UAV;

[0135] When the loop closure confidence is less than a preset threshold, the drone is controlled to repeatedly cruise in the current area in order to scan the scan data from different perspectives in the current area, and the loop closure confidence is recalculated by combining the scan data from different perspectives to make a loop closure confidence judgment.

[0136] When the loop closure confidence is greater than or equal to a preset threshold, global attitude graph optimization is performed.

[0137] Understandably, loopback, where the drone returns to an area it has previously traversed, identifies that it has returned to a place it has already visited, and can be used to correct map and pose errors and improve target detection quality.

[0138] In this disclosure, it is possible to determine whether to perform global attitude graph optimization based on loop closure confidence.

[0139] Specifically, when the system detects a loop closure candidate (i.e., in the current pose) When it observes features similar to a historical pose, it does not accept them directly, but instead conducts a confidence assessment.

[0140] Set at candidate pose Loop observations appear at [location] Define binary variables .

[0141] Use Bayes' theorem to update the probability of a true loop closure:

[0142] ;

[0143] in, This is the prior probability, which is the initial likelihood that the loop is true.

[0144] Let be the likelihood probability. It represents the probability that the currently observed data z (including feature matching consistency, geometric closure error, phase stability index, etc.) is likely to occur if the loop closure is true. These indices together form a scoring function; the higher the score, the greater the likelihood probability.

[0145] This represents the posterior probability, which is the current confidence level for the loop to be true after considering all observed evidence.

[0146] The system can set a confidence threshold. Posterior probability The calculation results will determine the subsequent behavior of the system:

[0147] For low confidence situations Instead of performing global map corrections, the area is marked as a "review area" and a repeated patrol (re-patrol) is triggered. The purpose is to allow the drone to fly over the suspicious area again to collect more data from different perspectives in order to obtain more reliable matching evidence. During the review phase, the data collected is only optimized and aligned within a local submap to avoid mismatches contaminating the global map.

[0148] For high confidence situations At this point, the system considers the loop closure matching to be highly reliable, and therefore performs global pose graph optimization. By utilizing this reliable constraint to correct the pose and map errors accumulated throughout the SLAM process, high-precision loop closure can be achieved.

[0149] This graded correction method reduces the risk of mismatch amplification error and is linked to the imaging quality threshold. When the phase is stable or the overlap is below the threshold, it prioritizes verification rather than forced image optimization. Figure 4 A detailed state machine is provided.

[0150] The methods provided by the embodiments of this disclosure have been fully described above. For ease of understanding, the following will be combined with... Figure 5 The complete implementation process of this disclosure is summarized.

[0151] Figure 5 A schematic diagram of the overall architecture of an embodiment of this disclosure is shown. (As shown) Figure 5As shown, the system in this disclosure follows a three-layer division of labor. The perception layer is responsible for millimeter-wave echo processing, feature point construction, IMU fusion, and trajectory estimation; the planning layer performs coverage band generation, step size and overlap evaluation, Bezier smoothing, and attitude-phase window coordination within the curved grid coordinate system; the execution layer sends the optimized velocity, attitude, and gimbal pointing to the flight controller, and uses a hybrid strategy of PID controller and LQR to ensure trajectory execution.

[0152] The system operates on a cyclical basis. First, the perception layer constructs short-term trajectories and local point clouds using radar odometers and IMUs. Second, the planning layer generates coverage bands of equal arc length and overlap within the parameter domain based on a curved grid coordinate system, performs Bezier smoothing and attitude constraint shaping on the trajectory, and provides the maximum angular velocity and scan rate based on phase window constraints. Third, the execution layer performs closed-loop control based on the desired wall distance and attitude profile. When a loop closure candidate appears, a probabilistic map update is initiated. If the confidence level is insufficient, a re-cruise is triggered to improve local robustness; if the confidence level is sufficient, global optimization is performed.

[0153] Based on the same inventive concept, this disclosure also provides a drone flight control device, as shown in the following embodiment. Since the principle by which this device embodiment solves the problem is similar to that of the above-described method embodiment, the implementation of this device embodiment can refer to the implementation of the above-described method embodiment, and repeated details will not be elaborated further.

[0154] Figure 6 This diagram illustrates the structure of a drone flight control device according to an embodiment of the present disclosure, as shown below. Figure 6 As shown, the UAV flight control device 600 is used for pipe inspection and includes: a setup module 601, a determination module 602, a fitting module 603, and a control module 604.

[0155] Module 601 is established to construct a family of skeleton curves based on the topological features of the pipe-pile surface structure, and to establish a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters.

[0156] The determination module 602 is used to determine the step distance required for the coverage of the pipe array curved surface and the constraints of the UAV attitude control by combining the performance parameters of the millimeter-wave radar on the UAV with the preset coverage requirements.

[0157] The fitting module 603 is used to generate a covering strip in the coordinate system of the curved surface grid based on the step size, and fit the covering strip into a smooth trajectory, wherein the smooth trajectory satisfies the constraint conditions and the minimum turning radius requirements;

[0158] The control module 604 is used to control the UAV to fly along a smooth trajectory and to inspect the pipe structure during flight, and output the inspection results.

[0159] In some embodiments, the UAV flight control device 600 further includes a loop closure detection module (not shown in the figure), which is used to calculate the loop closure confidence based on the likelihood between the current scan data and the historical scan data of the UAV; when the loop closure confidence is less than a preset threshold, the UAV is controlled to repeatedly cruise in the current area in order to scan the scan data from different perspectives in the current area, and the loop closure confidence is recalculated in combination with the scan data from different perspectives to make a loop closure confidence judgment; when the loop closure confidence is greater than or equal to the preset threshold, global attitude map optimization is performed.

[0160] In some embodiments, a module 601 is established to extract the central axis of the surface structure and generate a family of skeleton curves using a spline fitting method; local surface patches are constructed using the equal arc length parameter of the skeleton curve as the parameter along the skeleton direction and the distance perpendicular to the skeleton curve as the normal offset parameter; and continuous splicing conditions are applied to adjacent local surface patches to form an overall surface grid coordinate system.

[0161] In some embodiments, the determining module 602 is used to determine the projection width of the radar beam on the curved surface based on the geometric relationship between the radar pose and the curved surface and the half-power width of the geometric radar; determine the coverage step size perpendicular to the skeleton direction according to the preset minimum coverage overlap and projection width; and transform the phase stability constraint into a constraint condition based on the maximum allowable phase window, radar wavelength and equivalent lever arm.

[0162] In some embodiments, the fitting module 603 is used to generate parallel coverage strips in the curved surface grid coordinate system according to the coverage step size perpendicular to the skeleton direction; wherein, adjacent coverage strips are connected by a serpentine path strategy, and the end segments of the coverage strips are provided with redundant overlapping areas; each coverage strip is fitted as a piecewise curve; with the goal of suppressing curve curvature and jerk, combined with constraints and minimum turning radius requirements, each piecewise curve is smoothed and optimized to generate a smooth trajectory.

[0163] In some embodiments, the fitting module 603 is used to fit each coverage strip as a piecewise cubic Bézier curve using the following formula:

[0164] ;

[0165] in, For Bézier curves, It is the standardized arc length parameter along the curve. These are all control points in a cubic Bézier curve.

[0166] In some embodiments, the fitting module 603 is used to define the smoothing cost function:

[0167] ;

[0168] in, For smoothing cost function, This is the curvature weighting coefficient. For jerk weighting coefficient, To accelerate, For curvature, , for The first derivative, for The second derivative;

[0169] By adjusting the position of the control points, the smoothing cost function is minimized, and the generated smooth curve satisfies the constraints and minimum turning radius requirements, thus achieving smooth optimization of the piecewise curve.

[0170] In some embodiments, the UAV flight control device 600 further includes an optimization module (not shown) for defining a joint cost function for coverage gaps, overlap deviations, wall distance errors, phase errors, and energy consumption:

[0171] ;

[0172] in, For the joint cost function, For parameter field path, Used to measure the difference between actual overlap and target overlap. Areas not covered by penalties This is the wall distance error. This is a phase out-of-bounds term. For phase window, For the maximum allowed phase window, To control the input, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item;

[0173] During the flight of the UAV, the flight parameters of the UAV are optimized in a finite time domain based on the model predictive control algorithm to minimize the joint cost function and satisfy the preset constraints.

[0174] In some embodiments, the establishment module 601 is used to identify openings, corners and attached component areas in the curved structure and mark them as impassable areas in the curved grid coordinate system.

[0175] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0176] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, which may be a readable signal medium or a readable storage medium. A program product capable of implementing the methods described above is stored thereon. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code, which, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0177] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

Claims

1. A method for controlling the flight of an unmanned aerial vehicle (UAV), characterized in that, The method, applied to pipeline inspection, includes: Based on the topological features of the pipe-pile surface structure, a family of skeleton curves is constructed, and a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters is established. Based on the performance parameters of the millimeter-wave radar on the UAV and the preset coverage requirements, the step distance required for the pipe array curved surface coverage and the constraints of the UAV attitude control are determined. Within the curved mesh coordinate system, parallel coverage strips are generated with a coverage step perpendicular to the skeleton direction; adjacent coverage strips are connected using a serpentine path strategy, and redundant overlapping areas are set at the ends of the coverage strips. Each coverage strip is fitted as a piecewise curve; With the goal of suppressing curve curvature and jerk, and in conjunction with the constraints and minimum turning radius requirements, each segmented curve is smoothed and optimized to generate a smooth trajectory; wherein the smooth trajectory satisfies the constraints and minimum turning radius requirements. Control the drone to fly along a smooth trajectory and inspect the pipe structure during flight, outputting the inspection results.

2. The method according to claim 1, characterized in that, The method further includes: Calculate the loop closure confidence based on the likelihood between the current scan data and historical scan data of the UAV; When the loop closure confidence is less than a preset threshold, the drone is controlled to repeatedly cruise in the current area in order to scan the scan data from different perspectives in the current area, and the loop closure confidence is recalculated in combination with the scan data from different perspectives to make a loop closure confidence judgment. When the loop closure confidence is greater than or equal to a preset threshold, global attitude graph optimization is performed.

3. The method according to claim 1, characterized in that, The topological features based on the pipe-row surface structure are used to construct a family of skeleton curves, and a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters is established, including: Extract the central axis of the curved surface structure and generate a family of skeleton curves using spline fitting; Using the equal arc length parameter of the skeleton curve as the parameter along the skeleton direction, and the distance perpendicular to the skeleton curve as the normal offset parameter, a local surface patch is constructed. Apply continuous splicing conditions to adjacent local surface patches to form an overall surface grid coordinate system.

4. The method according to claim 1, characterized in that, The process of combining the performance parameters of the millimeter-wave radar on the UAV with preset coverage requirements determines the step distance required for the curved surface coverage of the pipe array and the constraints for UAV attitude control, including: Based on the geometric relationship between the radar pose and the curved surface, and combined with the radar's half-power width, the projection width of the radar beam on the curved surface is determined. Based on the preset minimum coverage overlap and the projection width, determine the coverage step size perpendicular to the skeleton direction; Based on the maximum allowable phase window, radar wavelength, and equivalent lever arm, the phase stability constraint is transformed into the aforementioned constraint condition.

5. The method according to claim 1, characterized in that, The step of fitting each coverage strip into a piecewise curve includes: Each coverage strip is fitted with a piecewise cubic Bézier curve using the following formula: ； in, For Bézier curves, It is the standardized arc length parameter along the curve. All of these are control points in a cubic Bézier curve.

6. The method according to claim 5, characterized in that, The process aims to suppress curve curvature and jerk, and, in conjunction with the constraints and minimum turning radius requirements, performs smoothing optimization on each segmented curve to generate a smooth trajectory, including: Define the smoothing cost function: ； in, For smoothing cost function, This is the curvature weighting coefficient. For jerk weighting coefficient, To accelerate, For curvature, , for The first derivative, for The second derivative; By adjusting the position of the control points, the smoothing cost function is minimized, and the generated smooth curve satisfies the constraints and minimum turning radius requirements, thereby achieving smooth optimization of the piecewise curve.

7. The method according to claim 1, characterized in that, The method further includes: Define a joint cost function for coverage gap, overlap deviation, wall distance error, phase error, and energy consumption: ； in, For the joint cost function, For parameter field path, Used to measure the difference between actual overlap and target overlap. Areas not covered by penalties This is the wall distance error. This is a phase out-of-bounds term. For phase window, For the maximum allowed phase window, To control the input, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item, for The weight coefficient of the item; During the flight of the UAV, the flight parameters of the UAV are optimized in a finite time domain based on the model predictive control algorithm to minimize the joint cost function and satisfy the preset constraints.

8. The method according to any one of claims 1 to 7, characterized in that, After constructing a family of skeleton curves based on the topological features of the pipe-pile surface structure and establishing a surface mesh coordinate system including parameters along the skeleton direction and normal offset parameters, the method further includes: Identify openings, corners, and attached component areas in the curved surface structure and mark them as impassable areas in the curved surface grid coordinate system.

9. A flight control device for unmanned aerial vehicles (UAVs), characterized in that, The device, used for pipeline inspection, includes: A module is established to construct a family of skeleton curves based on the topological features of the pipe-pile surface structure, and to establish a surface mesh coordinate system containing parameters along the skeleton direction and normal offset parameters. The determination module is used to combine the performance parameters of the millimeter-wave radar on the UAV with the preset coverage requirements to determine the step distance required for the pipe array curved surface coverage and the constraints of the UAV attitude control. The fitting module generates parallel coverage strips within the curved surface mesh coordinate system, with a coverage step size perpendicular to the skeleton direction. Adjacent coverage strips are connected using a serpentine path strategy, and redundant overlapping areas are provided at the ends of the coverage strips. Each coverage strip is fitted as a piecewise curve. Aiming to suppress curve curvature and jerk, and combining the aforementioned constraints and minimum turning radius requirements, each piecewise curve is smoothed and optimized to generate a smooth trajectory. The smooth trajectory satisfies the aforementioned constraints and minimum turning radius requirements. The control module is used to control the UAV to fly along a smooth trajectory and to inspect the pipe structure during flight, outputting the inspection results.

Images