Unmanned aerial vehicle recovery control system based on duct motors
By using image processing-based airflow disturbance recognition and PID control model, the speed of the ducted motor is dynamically adjusted, solving the attitude control lag problem caused by airflow asymmetry during UAV recovery and achieving reliable and efficient autonomous recovery of the UAV.
Patent Information
- Application Number
- CN202511627645.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-07
- Publication Date
- 2026-02-03
AI Technical Summary
During the recovery process of a drone in a low-pressure suction zone, the thrust disturbance caused by airflow asymmetry leads to attitude control lag, resulting in landing failure or deviation. Existing inertial device feedback control cannot effectively identify airflow disturbance signals.
Images are acquired using a downward-looking camera, the aerodynamic reflection zone is extracted, and linear fitting and Fourier transform processing are performed to calculate the vortex asymmetry coefficient. Combined with a PID control model, the speed of the ducted motor is dynamically adjusted to counteract airflow disturbances.
It enables real-time perception and feedforward control of airflow disturbances during UAV recovery, reducing recovery errors and improving the stability and robustness of soft landing.
Smart Images
Figure CN121454898A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of unmanned aerial vehicle recovery control, more particularly, it relates to a unmanned aerial vehicle recovery control system based on ducted motor. BACKGROUND
[0002] Currently, in the field of unmanned aerial vehicle recovery control, using the ducted motor array arranged on the top of the hangar to realize low-pressure suction recovery is an important means to improve the recovery accuracy and automation. This method forms a stable local low-pressure area above the hangar through the duct exhaust, so that the unmanned aerial vehicle obtains additional downward pressure when landing, and realizes soft adsorption landing.
[0003] However, in actual deployment, limited by platform structure, boundary disturbance and environmental wind field changes, the above-mentioned low-pressure area is often difficult to maintain ideal symmetry in spatial distribution. Especially when the suction of one side of the wind machine is stronger, or there are structural gaps and lateral leakage on the edge of the hangar platform, the phenomenon of stable suction flow on one side and poor return flow or forming a backflow vortex on the other side often occurs. This uneven airflow distribution will directly act on the bottom of the unmanned aerial vehicle body, causing asymmetric thrust disturbance. In the process of the unmanned aerial vehicle gradually landing and approaching the low-pressure suction area, the thrust disturbance shows small amplitude and periodic rolling or pitching oscillation, and the frequency is usually in the range of 1 to 3 Hz. Although the amplitude is not large, due to the high position control accuracy, such perturbations may cause the oscillation response of the attitude controller in the flight control system, and finally lead to landing failure, deviation from the center or recovery failure.
[0004] The traditional unmanned aerial vehicle flight control system mainly relies on gyroscopes, accelerometers and other inertial devices to realize real-time sensing of the body attitude and motion changes. However, the thrust disturbance is often difficult to be reflected in the inertial navigation system in a short time scale. When the IMU detects the attitude change, the system has entered the disturbance feedback link, and the compensation control has obvious hysteresis, especially in the stage when the unmanned aerial vehicle is about to land, the position accuracy is the highest. The delay may cause the system to misjudge and miss the most effective suppression window.
[0005] During the landing process of the unmanned aerial vehicle, the downward-looking camera is usually in working state, which is used for visual positioning, alignment or terrain recognition. In some specific conditions, such as the platform surface being metal or hard board, the flow structure generated by the duct suction will form a clear dynamic image signal in the ground dust particles. Especially when only one side forms a backflow vortex, the dust particles are sucked away at high speed, and the exposed area presents a typical wedge-shaped bright spot in the image, with the top angle aligned with the backflow direction, and the wedge tip presents a periodic oscillation trend along the time axis. This phenomenon reflects the asymmetry of airflow distribution, which is highly related to the possible attitude disturbance.
[0006] However, this phenomenon has not been collected or identified by the traditional flight control system, resulting in the important disturbance warning signal carried by it being ignored for a long time. Relying only on inertial devices for feedback control cannot solve the hidden chain of attitude micro-vibration caused by the suction flow disturbance source, control lag, and recovery failure.
[0007] Therefore, how to extract the indirect signs of this asymmetric flow from the unmanned aerial vehicle image and convert its structural features into control parameters, and to adjust the fan output in advance to actively weaken the disturbance source when the unmanned aerial vehicle has not yet produced obvious errors, is a technical problem to be solved at present. SUMMARY
[0008] The present application provides a kind of unmanned aerial vehicle recovery control system based on ducted motor, solve the technical problems in the background art.
[0009] The present application provides a kind of unmanned aerial vehicle recovery control system based on ducted motor, comprising: Critical determination unit, for determining critical orientation difference based on the length and width of target hangar; Image acquisition unit, for obtaining orthographic image by downward-looking camera at fixed sampling frequency, to extract aerodynamic reflection area based on orthographic image; Fitting unit, for linear fitting based on aerodynamic reflection area, to obtain first straight line and second straight line, and determine orientation angle, orientation angle difference value and orientation angle mean value of first straight line and second straight line in turn; Symmetry processing unit, for determining vortex asymmetry coefficient according to orientation angle difference value and critical orientation difference; Frequency processing unit, for extracting M orientation angle mean values within preset time window, and obtaining reflection area oscillation frequency and unmanned aerial vehicle jitter frequency by Fourier transform processing; Recovery control unit, for differential PID recovery control management of unmanned aerial vehicle according to reflection area oscillation frequency, unmanned aerial vehicle jitter frequency and vortex asymmetry coefficient, to realize soft landing control of unmanned aerial vehicle.
[0010] Further, based on the length and width of target hangar, the critical orientation difference is determined, comprising: Read the length and width of target hangar opening; Take the inverse tangent value of the ratio of width to length as the critical orientation difference.
[0011] Further, based on the orthographic image, the aerodynamic reflection area is extracted, comprising: Obtain the original image through the downward-looking camera; Convert the original image through the preset inverse perspective transformation matrix to form the orthographic image; Extract the orthographic image in the preset previous time period at the current time, and calculate the mean image; A difference image is formed by calculating the difference between the orthographic image and the mean image at the current time; The difference image is grayed and a gray threshold is set to form a first mask by binary classification; wherein the first mask is a binary matrix, and element 1 represents a candidate reflection element, and other elements are 0; The morphological closing processing is performed on the first mask to form an initial aerodynamic reflection area; The K polar rays are generated with the centroid of the initial aerodynamic reflection area as the polar coordinate origin; wherein the polar angle range of the polar ray is , and the polar angle interval of adjacent polar rays is ; For the i-th polar ray, the initial aerodynamic reflection area is sampled from the centroid of the initial aerodynamic reflection area, and if the sampling value changes from 1 to 0, the Euclidean distance of the element to the centroid is extracted ; For the i-th polar ray, the rhythm value is calculated as follows: wherein, represents the Euclidean distance of the i+1-th polar ray, the Euclidean distance of the i-1-th polar ray; If the rhythm value of at least polar rays is greater than the preset rhythm threshold, the initial aerodynamic reflection area is marked as valid, otherwise the gray threshold is adjusted until at least polar rays have a rhythm value greater than the preset rhythm threshold, and the aerodynamic reflection area is obtained.
[0012] Further, based on the aerodynamic reflection area, a straight line fitting is performed to obtain a first straight line and a second straight line, and the orientation angle, the orientation angle difference value and the orientation angle mean value of the first straight line and the second straight line are determined in turn, including: The boundary pixels of the aerodynamic reflection area are sequentially extracted in the orthographic image to obtain an outer contour; The curvature value of the i-th pixel in the outer contour is calculated by extracting a preset number of pixels before and after the i-th pixel in the outer contour as the center; The pixel with the maximum curvature value in the outer contour is extracted as a segmentation pixel; The outer contour is divided into a first pixel set and a second pixel set by the segmentation pixel; The first straight line and the second straight line are fitted respectively for the first pixel set and the second pixel set; The inverse tangent values of the slopes of the first straight line and the second straight line are obtained to obtain the corresponding orientation angles; The difference between the orientation angles of the first straight line and the second straight line is obtained to obtain the orientation angle difference value; The average of the orientation angles of the first straight line and the second straight line is calculated to obtain the orientation angle mean value.
[0013] Furthermore, based on the orientation angle difference and the critical orientation difference, the vortex asymmetry coefficient is calculated, including: The angle residual is obtained by subtracting the orientation angle difference from the critical orientation difference. The ratio of the angular residual to the preset normalized residual is used as the vortex asymmetry coefficient.
[0014] Furthermore, the mean values of M orientation angles within a preset time window are extracted and processed using Fourier transform to obtain the oscillation frequency of the reflector zone and the jitter frequency of the UAV, including: Extract the mean values of M orientation angles within a preset time window to form a sequence of mean orientation angles; Calculate the average value of the orientation angle mean sequence to obtain the DC component; The mean value of the orientation angle sequence is reduced by the DC component to obtain a zero-mean sequence. Construct a Hanning window function hn of length M, n∈ Where n is 1 or M, the Hanning window function takes the value of 0; when n is 1 or M, the value of n is M. At that time, the Hanning window function reaches its maximum value; The zero-mean sequence is weighted sequentially using the Hanning window function to obtain a preprocessed sequence; Determine the amplitude spectrum for the m-th unit in the preprocessed sequence, including: Perform Fourier transform on the m-th unit in the preprocessed sequence to obtain the complex spectrum, and use the modulus of the complex spectrum as the amplitude spectrum of the m-th unit; The product of the ratio of the fixed sampling frequency and M and m is used as the frequency of the m-th unit; If the amplitude spectrum of the m-th unit is the largest, then the frequency of the m-th unit is taken as the oscillation frequency of the reflection zone, and half of the oscillation frequency of the reflection zone is taken as the jitter frequency of the UAV according to the prior proportional relationship.
[0015] Furthermore, differential PID recovery control management is implemented for the drones, including: The initial differential error is calculated based on the drone's jitter frequency, and then weighted by a vortex asymmetry coefficient to form the differential error. ; Determine the system's natural frequency Preset damping ratio ζ and total moment of inertia ;in, This represents the drone's jitter frequency and calculates the weight of the proportional term in the PID gain. Weight of integral terms and differential term weights ;in, , and ; Based on differential error Combine proportional term weight , integral term weight and differential term weight Calculate the proportional term error , integral term error and differential term error of PID gain respectively Wherein, denotes the integral term error at time , denotes the differential error at time , denotes the fixed sampling frequency Based on the proportional term error, integral term error and differential term error, the differential speed command is synthesized.
[0016] Further, the differential PID recovery control management of the unmanned aerial vehicle also includes: Determine the reference speed of the left and right duct motors of the target unmanned aerial vehicle ; The sum of the reference speed and the differential speed command is taken as the target speed of the left duct motor The difference between the reference speed and the differential speed command is taken as the target speed of the right duct motor
[0017] Further, the differential PID recovery control management of the unmanned aerial vehicle also includes: According to the vortex asymmetry coefficient, the sinking speed of the target unmanned aerial vehicle is controlled , as follows: Wherein, denotes the initial sinking speed of the target unmanned aerial vehicle , denotes the vortex asymmetry coefficient , and denotes the sinking weight
[0018] The beneficial effect of the present application is that: by introducing a disturbance area recognition mechanism based on image brightness and structural changes, combined with the low pressure suction flow generated by the ducted motor, the real-time perception and feedforward control of the air flow disturbance source in the unmanned aerial vehicle recovery process are realized. The system can extract the asymmetric distribution information in the suction flow field according to the orientation angle change of the high-light area edge appearing in the image, the boundary rhythm characteristics and the angle swing frequency, and further construct a difference PID control model to dynamically adjust the left and right rotation speed and sinking rate of the ducted motor. Before the unmanned aerial vehicle has not occurred attitude deviation, that is, the identification and response of the disturbance trend are completed, the recovery error caused by uneven air flow is significantly reduced, the stability and robustness of soft landing are improved, and reliable and efficient unmanned aerial vehicle autonomous recovery control is realized. BRIEF DESCRIPTION OF DRAWINGS
[0019] Fig. 1 is a module diagram of the present application; Fig. 2 is a flowchart of the present application; Fig. 3 is a recovery control flowchart of the unmanned aerial vehicle of the present application. DETAILED DESCRIPTION
[0020] The subject matter described herein will now be discussed with reference to example implementations. It should be understood that the discussion of these implementations is merely meant to provide a better understanding of the subject matter described herein and can include changes, modifications, additions or omissions of the functions and arrangements of the elements discussed without departing from the scope of the present description. Various examples can omit, substitute or add various procedures or components as appropriate, and also can vary from example to example. In addition, features described in relation to some examples can also be combined in other examples.
[0021] As shown in Figs. 1-3 , a ducted motor-based unmanned aerial vehicle recovery control system includes: A critical determination unit is configured to determine a critical orientation difference based on the length and width of a target hangar. An image acquisition unit is configured to acquire an orthographic image by a downward-looking camera at a fixed sampling frequency to extract a aerodynamic reflection area based on the orthographic image. A fitting unit is configured to perform linear fitting based on the aerodynamic reflection area to obtain a first straight line and a second straight line, and sequentially determine an orientation angle, an orientation angle difference and an orientation angle mean value of the first straight line and the second straight line. A symmetry processing unit is configured to determine a vortex asymmetry coefficient according to the orientation angle difference and the critical orientation difference. A frequency processing unit is configured to extract M orientation angle mean values within a preset time window, and perform Fourier transform processing to obtain a reflection area swing frequency and an unmanned aerial vehicle jitter frequency. The recovery control unit is used for carrying out differential PID recovery control management on the unmanned aerial vehicle according to the reflection area swing frequency, the unmanned aerial vehicle jitter frequency and the vortex asymmetry coefficient, so as to realize the soft landing control of the unmanned aerial vehicle.
[0022] In one embodiment of the present application, the critical orientation difference is determined based on the length and width of the target hangar, comprising: reading the length and width of the target hangar opening; taking the inverse tangent value of the ratio of the width to the length as the critical orientation difference.
[0023] It should be noted that the length and width of the target hangar opening are read. The hangar opening is the target area for the recovery of the unmanned aerial vehicle, and its geometric size (length and width) directly determines the spatial constraint of the airflow entering the hangar in the ideal state, that is, the length usually corresponds to the longer side of the hangar opening, and the width corresponds to the shorter side, and the proportional relationship between the two reflects the rectangular shape characteristics of the hangar opening. The inverse tangent value of the ratio of the width to the length is taken as the critical orientation difference. The essence of this calculation is to convert the geometric proportion of the hangar opening into an angle parameter. Geometrically, the ratio of the width to the length reflects the width-to-length ratio of the hangar opening, and the inverse tangent operation can convert the acute angle of the right triangle corresponding to the ratio (the opposite side is the width and the adjacent side is the length) into an angle value (critical orientation difference). The critical orientation difference objectively reflects the geometric orientation characteristics of the hangar opening itself, and can be used as a reference angle that the aerodynamic reflection area should follow in the ideal state (when the airflow is completely symmetrical).
[0024] In one embodiment of the present application, the aerodynamic reflection area is extracted based on the orthographic image, comprising: obtaining an original image through a downward-looking camera; converting and processing the original image through a preset inverse perspective transformation matrix to form an orthographic image; extracting the orthographic image in a preset previous time period at the current time and calculating a mean image; calculating the difference between the orthographic image at the current time and the mean image to form a difference image; graying the difference image and setting a gray threshold to form a first mask in a binary classification; wherein the first mask is a binary matrix, and element 1 represents a candidate reflection element, and the rest of the elements are 0; performing morphological closing processing on the first mask to form an initial aerodynamic reflection area; taking the centroid of the initial aerodynamic reflection area as the polar coordinate origin to generate K polar rays; wherein the polar angle range of the polar ray is , and the polar angle interval of adjacent polar rays is ; for the i-th polar ray, sampling the initial aerodynamic reflection area from the centroid of the initial aerodynamic reflection area outward, if the sampling value changes from 1 to 0, the Euclidean distance of the element to the centroid is extracted ; The rhythm value of the i-th polar ray is calculated as follows: wherein, represents the Euclidean distance of the i+1-th polar ray, the Euclidean distance of the i-1-th polar ray; If the rhythm value of at least polar rays is greater than a preset rhythm threshold, the initial aerodynamic reflection region is marked as valid, otherwise the gray threshold is adjusted until at least polar rays has a rhythm value greater than the preset rhythm threshold, and the aerodynamic reflection region is obtained.
[0025] It should be noted that after collecting the original image below the unmanned aerial vehicle by the downward-looking camera, the original image needs to be converted by a preset inverse perspective transformation matrix to obtain the orthographic image. Thus, the perspective distortion of the original image caused by the shooting angle (tilt) is eliminated, the image is restored to a plane view angle perpendicular to the ground, and it is ensured that the subsequent analysis of the shape of the ground airflow region is more consistent with the actual physical space.
[0026] All orthographic images in a preset previous time period before the current time are extracted, the mean value of these images is calculated to obtain a mean value image (background average state in this time period); and the pixel difference between the orthographic image at the current time and the mean value image is calculated to form a difference image. Since the aerodynamic reflection region is a dynamic region (bright spot formed by dust particles being sucked) generated by airflow disturbance, its brightness or shape in the image will change over time, while the background average state is relatively stable. The difference between the current time and the background average state can be highlighted by difference operation, thereby preliminarily locating the dynamic change region.
[0027] The difference image is subjected to grayscale processing, and a grayscale threshold is set: the pixels with a grayscale value higher than the grayscale threshold are determined as candidate reflection elements, and the rest are background. Thus, a first mask is formed, which is a binary matrix, wherein the positions with an element of 1 correspond to the candidate reflection elements, and the positions with an element of 0 correspond to the background. Thus, by grayscale threshold segmentation, the pixels possibly belonging to the aerodynamic reflection region are preliminarily screened out from the difference image, and the dynamic region and the background are preliminarily separated.
[0028] Morphological closing processing (by the operation of dilating and then eroding) is performed on the first mask to form an initial aerodynamic reflection region. Since the difference operation and threshold segmentation may leave small noise points (such as isolated bright pixels), or there are small gaps in the candidate reflection elements, the morphological closing processing can fill in these gaps and eliminate small noise points, so that the candidate region forms a continuous and complete region shape.
[0029] It should be noted that the centroid of the initial aerodynamic reflection region is taken as the polar coordinate origin, and K polar rays are generated (the rays are uniformly distributed in a preset polar angle range, and the polar angle interval of adjacent rays is ). The role of the polar ray is to detect the boundary form from the region center to the periphery. For each polar ray, the mask value (1 or 0) of the initial aerodynamic reflection region is sampled point by point from the centroid outward. When the sampling value changes from 1 (inside the reflection region) to 0 (the external background), the point is the boundary point of the reflection region, and the Euclidean distance (reflecting the spatial position of the boundary point) of the boundary point to the centroid of the initial aerodynamic reflection region is extracted. For the i-th polar ray, the “rhythm value” is calculated by the formula: the rhythm value is the absolute value of the difference between the boundary distances of the ray and the adjacent two rays (i-1, i+1), reflecting the change rule of the adjacent boundary points. The real aerodynamic reflection region has a stable dynamic structure due to airflow disturbance, and the boundary change will present a certain rhythm (the rhythm value is larger); and the boundary change of the noise region is irregular (the rhythm value is smaller). If the rhythm value of at least polar rays is greater than a preset rhythm threshold, it is indicated that the region boundary has a stable rhythm, and is marked as an effective initial aerodynamic reflection region.
[0030] In an embodiment of the present application, a straight line fitting is performed based on the aerodynamic reflection region to obtain a first straight line and a second straight line, and an orientation angle, an orientation angle difference value and an orientation angle average value of the first straight line and the second straight line are determined in turn, including: Boundary pixels of the aerodynamic reflection region are sequentially extracted in the orthographic image to obtain an outer contour; An i-th pixel in the outer contour is taken as a center, and a preset number of pixels in the outer contour are extracted before and after the i-th pixel to calculate a curvature value of the i-th pixel; A pixel with the maximum curvature value in the outer contour is extracted as a segmentation pixel; The outer contour is divided into a first pixel set and a second pixel set by the segmentation pixel; A first straight line and a second straight line are fitted respectively for the first pixel set and the second pixel set; Respective arctangent values of slopes of the first straight line and the second straight line are taken to obtain corresponding orientation angles; The orientation angles of the first straight line and the second straight line are subtracted to obtain an orientation angle difference value; An average value of the orientation angles of the first straight line and the second straight line is calculated to obtain an orientation angle average value.
[0031] It should be noted that, with the i-th pixel in the outer contour as the center, a preset number of pixels are extracted forward and backward (5 pixels each, by the discrete curvature calculation method), and the curvature value of the pixel is calculated by the coordinate change of these pixels (the greater the curvature value, the more significant the bending degree of the contour at this position). The outer contour of the aerodynamic reflection zone is determined by the airflow pattern, and when the airflow is asymmetric or vortex is formed, the contour will have a significant corner at a certain position. By calculating the curvature value, the most significant bending pixel in the contour can be located, which usually corresponds to the key turning point of the airflow pattern and is an ideal node for segmenting the contour.
[0032] The pixel with the maximum curvature value in the outer contour is extracted as the segmentation pixel, and the outer contour is divided into a first pixel set and a second pixel set based on the segmentation pixel. The outer contour of the aerodynamic reflection zone usually presents a non-single direction pattern due to airflow disturbance (one side extends along the airflow direction, and the other side is deflected due to vortex), and direct fitting of a straight line will result in errors. Splitting the contour with the segmentation pixel with the maximum curvature can convert the complex contour into two relatively regular line segments, making the subsequent straight line fitting more accurate and reflecting the corresponding airflow direction.
[0033] The first pixel set and the second pixel set are respectively fitted with straight lines by the least squares method to obtain a first straight line and a second straight line. Straight line fitting can abstract the overall direction of the contour segment into straight line parameters (slope and intercept), eliminating the influence of local minor fluctuations. The two straight lines correspond to the dominant directions of the two parts of the aerodynamic reflection zone, and the difference in their directions directly reflects the degree of airflow asymmetry.
[0034] The inverse tangent values of the slopes of the first straight line and the second straight line are calculated respectively to obtain the corresponding orientation angles. The slope reflects the inclination of the straight line, but the value is affected by the coordinate scale; while the orientation angle can standardize the direction and facilitate the comparison of the directions of different straight lines. For example, a straight line with a slope of 1 corresponds to an orientation angle of 45 degrees, and a straight line with a slope of 0 corresponds to an orientation angle of 0 degrees, which intuitively reflects the spatial direction of the straight line.
[0035] The difference between the orientation angles of the first straight line and the second straight line is obtained. The orientation angle difference directly reflects the direction deviation of the two straight lines: the smaller the difference, the closer the directions of the two contour parts, corresponding to a more symmetric airflow; the larger the difference, the more significant the deviation of the two directions, corresponding to a stronger airflow asymmetry.
[0036] The average of the orientation angles of the first straight line and the second straight line is calculated to obtain the average orientation angle. The average orientation angle represents the dominant direction of the aerodynamic reflection zone as a whole, and can reflect the overall trend of the airflow (the overall direction extends in a certain direction, corresponding to the main airflow direction). The value changes over time (periodic oscillation), which is the basis for subsequent extraction of the reflection zone oscillation frequency.
[0037] In one embodiment of the present application, the vortex asymmetry coefficient is calculated according to the orientation angle difference value and the critical orientation difference, comprising: The orientation angle difference value is subtracted from the critical orientation difference to obtain an angle residual; The ratio of the angle residual and a preset normalized residual is taken as the vortex asymmetry coefficient.
[0038] It should be noted that the orientation angle difference value reflects the actual direction difference of the two fitted straight lines in the aerodynamic reflection area, which is caused by airflow disturbance (such as vortex, lateral leakage); and the critical orientation difference is an ideal value determined based on the hangar geometric size, that is, it represents the direction difference reference of the two straight lines in the aerodynamic reflection area when the airflow is completely symmetrical. The difference value (angle residual) of the two directly reflects the deviation degree of the actual airflow direction difference from the ideal symmetrical state: if the angle residual is 0, it means that the actual airflow is close to the ideal symmetrical state; if the absolute value of the angle residual is larger, it means that the actual airflow deviates from the ideal state more significantly due to vortex or asymmetric disturbance.
[0039] The ratio of the angle residual and a preset normalized residual is taken as the vortex asymmetry coefficient. The numerical value of the angle residual is affected by the specific scene (such as hangar size, airflow disturbance intensity), and direct use may cause the subsequent control logic to be not universal due to scale difference. The preset normalized residual is a reference residual value (maximum angle residual under typical disturbance) set according to the actual application scene, and through ratio operation, the angle residual can be converted into a dimensionless coefficient, that is, the vortex asymmetry coefficient. The physical meaning of the vortex asymmetry coefficient is the proportion of the actual deviation degree relative to the reference deviation degree: the larger the absolute value of the coefficient, the more serious the airflow asymmetry; the coefficient is 0 or close to 0, indicating that the airflow is close to the symmetrical state.
[0040] In one embodiment of the present application, the M orientation angle means in the preset time window are extracted and processed by Fourier transform to obtain the reflection area swing frequency and the unmanned aerial vehicle jitter frequency, comprising: The M orientation angle means in the preset time window are extracted to form an orientation angle mean sequence; The average value of the orientation angle mean sequence is calculated to obtain a direct current component; The orientation angle mean sequence is de-meaned with the direct current component to obtain a zero-mean sequence; A Hanning window function hn, n∈ is constructed, where when n is 1 or M, the Hanning window function takes a value of 0, and when n is , the Hanning window function takes a maximum value; The zero-mean sequence is weighted by the Hanning window function to obtain a preprocessed sequence; The amplitude spectrum corresponding to the mth unit in the preprocessed sequence is determined, comprising: The Fourier transform is performed on the mth unit in the pretreatment sequence to obtain a complex spectrum, and the modulus value of the complex spectrum is taken as the amplitude spectrum of the mth unit; The product of the ratio of the fixed sampling frequency and M and m is taken as the frequency of the mth unit; If the amplitude spectrum of the mth unit has the maximum value, the frequency of the mth unit is taken as the reflection zone oscillation frequency, and half of the reflection zone oscillation frequency is taken as the unmanned aerial vehicle jitter frequency according to the prior proportional relationship.
[0041] It should be noted that the mean value of the orientation angle reflects the overall direction of the aerodynamic reflection zone, and the airflow oscillation (periodic shift of the vortex) causes the mean value to periodically change over time (when the orientation angle alternately increases and decreases, the mean value of the orientation angle alternately increases and decreases). The preset time window covers at least one complete oscillation period, M is the number of samples in the window (for example, if the window is 1 second and the sampling frequency is 10 Hz, then M = 10), and the sequence records the continuous data of the direction change in this period of time, providing a basis for subsequent frequency analysis.
[0042] The average value of the sequence of mean values of the orientation angle is calculated to obtain the direct current component (representing the constant offset of the overall sequence, such as the fixed trend of the overall reflection zone being biased in a certain direction); and the sequence is de-meaned using the direct current component to obtain a zero-mean sequence (each value in the sequence is subtracted by the direct current component). The direct current component only reflects the fixed direction offset and is irrelevant to the periodic oscillation; after de-meaning, the sequence only retains the part that fluctuates over time (i.e., the periodic change component), which can avoid the interference of the constant offset on the extraction of the oscillation frequency.
[0043] A Hanning window function hn with a length of M is constructed (n is the sequence index, ranging from 1 to M), where n is 0 when n is 1 or M, and n is maximum when n is The function is bell-shaped with high middle and low ends); the zero-mean sequence is weighted by the function one by one to obtain the pretreatment sequence. When Fourier transform is performed on a non-strictly periodic signal, frequency spectrum leakage (energy of a certain frequency spreading to adjacent frequencies) easily occurs, leading to frequency identification deviation. The Hanning window reduces the weight of the two ends of the sequence (the beginning and end of the fluctuation) and increases the weight of the middle (the stable fluctuation part), which can simulate the effect of periodic extension, reduce the interference of non-periodic components, make the energy of the periodic components in the subsequent spectrum more concentrated, and improve the accuracy of frequency extraction.
[0044] The corresponding amplitude spectrum and frequency of the mth unit (m is an index ranging from 1 to M) in the pretreatment sequence are determined: The Fourier transform is performed on the mth unit in the pretreatment sequence to obtain a complex spectrum (containing amplitude and phase information of the frequency), and the modulus value of the complex spectrum is taken as the amplitude spectrum of the unit, i.e., the amplitude spectrum reflects the intensity of the fluctuation at this frequency (the larger the amplitude, the more significant the periodicity at this frequency).
[0045] The frequency of the mth unit is calculated: the ratio of the fixed sampling frequency to M (i.e. the frequency resolution, representing the interval of adjacent frequencies in the frequency domain) is multiplied by m to obtain the corresponding frequency, which is the correspondence between the time domain sampling and the frequency in the frequency domain Fourier transform, ensuring that each unit corresponds to a unique fluctuation frequency.
[0046] If the amplitude spectrum of the mth unit has the maximum value, it indicates that the frequency corresponding to the unit is the most significant periodic component in the sequence, and it is taken as the oscillation frequency of the reflection region, i.e. the frequency directly reflects the frequency of the direction reciprocating change of the aerodynamic reflection region due to the airflow oscillation (periodic shift of vortex). According to the prior proportional relationship (based on the law of interaction between airflow and unmanned aerial vehicle, there is a stable proportion between reflection region oscillation and unmanned aerial vehicle jitter), half of the reflection region oscillation frequency is taken as the unmanned aerial vehicle jitter frequency, because the airflow oscillation will cause the unmanned aerial vehicle jitter through the thrust disturbance, and the jitter frequency is usually half of the airflow oscillation frequency, which can predict the periodic attitude disturbance frequency of the unmanned aerial vehicle.
[0047] In an embodiment of the present application, the differential PID recovery control management is performed on the unmanned aerial vehicle, comprising: The initial differential error is calculated based on the unmanned aerial vehicle jitter frequency, and the initial differential error is weighted based on the vortex asymmetry coefficient to form the differential error ; The system inherent frequency , the preset damping ratio ζ and the total moment of inertia are determined; wherein, represents the unmanned aerial vehicle jitter frequency, and the proportional term weight , the integral term weight and the differential term weight of the PID gain are calculated; wherein, , and ; Based on the differential error , the proportional term error , the integral term error and the differential term error of the PID gain are calculated respectively based on the proportional term weight , the integral term weight and the differential term weight ; wherein, represents the integral term error at time , represents the integral term error at time differential error, represents a fixed sampling frequency; The differential speed command is synthesized based on the proportional error, the integral error and the differential error.
[0048] It is to be noted that the initial differential error is calculated based on the UAV jitter frequency, and the initial differential error reflects the attitude deviation trend that may be caused by periodic jitter. The higher the jitter frequency, the greater the initial error. The differential error is obtained by weighting the initial differential error with the vortex asymmetry coefficient . The UAV jitter is caused by airflow disturbance, and the vortex asymmetry coefficient directly reflects the severity of the disturbance, that is, the greater the asymmetry coefficient, the more significant the influence of jitter on attitude. Through weighting, the differential error can reflect both the jitter frequency (periodicity of disturbance) and the asymmetry degree (intensity of disturbance): when the asymmetry is serious, the differential error is amplified, and the subsequent control adjustment is more aggressive; otherwise, the error is smaller, and the adjustment is more gentle.
[0049] In detail, the system natural frequency is a parameter calculated based on the UAV jitter frequency, reflecting the natural frequency characteristics of the attitude response of the UAV under the current airflow disturbance. The UAV jitter frequency is caused by airflow disturbance, and the system natural frequency converts the jitter frequency into a parameter of the adaptive control model. The higher the jitter frequency (the stronger the periodicity of airflow disturbance), the greater the system natural frequency, meaning that the control model needs a faster response speed to match the disturbance rhythm and avoid attitude deviation aggravation due to response lag.
[0050] The preset damping ratio ζ is a parameter preset according to the UAV recovery stability requirement (determined by the actual control scenario), to control the stability of UAV attitude adjustment. The greater the damping ratio, the more gentle the system response, which can avoid attitude oscillation caused by excessive adjustment, such as reciprocating swing caused by excessive correction. If the damping ratio is too small, the system may be too sensitive to disturbance, resulting in excessive adjustment amplitude. By presetting a suitable ζ, a balance can be achieved between fast response to disturbance and avoidance of attitude oscillation.
[0051] The total moment of inertia is a parameter reflecting the overall moment of inertia of the UAV, determined by the structure of the UAV itself, including the mass distribution, size, etc. The total moment of inertia measures the difficulty of changing the attitude of the UAV. The greater the total moment of inertia, the more difficult it is to change the attitude of the UAV (a greater moment is required to adjust the attitude); otherwise, it is easier to adjust. This parameter directly affects the calculation of the PID gain, that is, the greater the moment of inertia, the greater the control moment required to ensure that the adjustment command can effectively offset the airflow disturbance and achieve attitude correction.
[0052] It should be noted that the proportional term is used to address immediate response errors, and its weight is positively correlated with the natural frequency, damping ratio, and moment of inertia. This means that the higher the system response requirement and the greater the inertia, the stronger the proportional adjustment needs to be. The integral term is used to eliminate accumulated errors, and its weight is derived from the proportional term and the natural frequency, ensuring that the integral adjustment matches the system response speed and avoiding over-adjustment. The derivative term is used to suppress the rate of change of error, and its weight is equal to the moment of inertia. This means that the greater the inertia, the more necessary it is to suppress attitude change trends in advance through derivative adjustment.
[0053] It should be noted that, based on the differential error and the aforementioned weights, the proportional, integral, and derivative errors are calculated separately to decompose the control requirements from different dimensions: The proportional term error is directly proportional to the current differential error and is used for immediate response. That is, when the differential error increases, the proportional term error immediately increases, and an adjustment command is quickly output to counteract the current disturbance.
[0054] The integral term error is used to eliminate steady-state error. If the disturbance persists, the integral error will gradually increase to ensure that the control command continues to act until the error returns to zero.
[0055] The differential error is used for early suppression. That is, when the error increases rapidly, the differential error increases immediately, and the reverse adjustment command is output in advance to suppress further attitude deviation.
[0056] The differential speed command is generated based on the sum of the proportional, integral, and differential errors. This command directly determines the speed difference between the left and right ducted motors: when the command is positive, the left motor speed increases and the right motor speed decreases (to counteract leftward disturbances); when the command is negative, the opposite occurs to ensure attitude stability during UAV recovery.
[0057] In one embodiment of the present invention, differential PID recovery control management of the UAV further includes: Determine the reference speeds of the left and right ducted motors of the target UAV. ; Reference speed The sum of the differential speed command and the target speed of the left bypass motor is used as the target speed. Reference speed The difference between the speed command and the differential speed command is used as the target speed of the right duct motor.
[0058] It should be noted that the reference speed is the basic speed required for the drone to achieve stable descent in an ideal symmetrical airflow environment. At this speed, the left and right motors rotate at the same speed, generating balanced lift, allowing the drone to maintain a horizontal attitude and descend at a uniform speed. The setting of the reference speed needs to be matched with parameters such as the drone's own weight and initial descent speed to ensure sufficient lift to balance gravity, while maintaining the preset descent rhythm. It serves as a benchmark reference for subsequent dynamic adjustments.
[0059] The sum of the reference speed and the differential speed instruction is taken as the target speed of the left nacelle motor. The differential speed instruction is a dynamic adjustment quantity calculated based on airflow disturbance (vortex asymmetry coefficient, UAV jitter frequency), and the sign and size of the differential speed instruction directly reflect the disturbance direction and intensity. When the differential speed instruction is positive, the target speed of the left nacelle motor is higher than the reference speed, that is, the lift generated by the left motor increases at this time, which can offset the tendency of the UAV to tilt to the left due to disturbance; if the instruction is negative, the left motor speed is lower than the reference speed, which can offset the tendency to tilt to the right.
[0060] The difference between the reference speed and the differential speed instruction is taken as the target speed of the right nacelle motor. The target speed of the right motor forms a symmetrical adjustment with the left motor: when the differential speed instruction is positive, the speed of the right motor is lower than the reference speed (in contrast to the increase of the left motor), and the speed difference between the left and right motors generates a clockwise adjustment torque, which offsets the disturbance to the left; when the instruction is negative, the speed of the right motor is higher than the reference speed, and the speed difference generates a counterclockwise torque, which offsets the disturbance to the right. This one-increase and one-decrease speed adjustment can achieve effective attitude correction with the smallest speed change, avoiding new disturbance caused by large unilateral speed change.
[0061] In an embodiment of the present application, the differential PID recovery control management is performed on the UAV, and further comprises: According to the vortex asymmetry coefficient, the sinking speed of the target UAV is controlled As follows: Wherein, represents the initial sinking speed of the target UAV, represents the vortex asymmetry coefficient, represents the sinking weight.
[0062] It should be noted that the initial sinking speed is the preset descending speed of the UAV when the airflow is symmetrical (the vortex asymmetry coefficient is 0), which is used as a basic reference value to ensure the recovery efficiency under normal circumstances; the vortex asymmetry coefficient reflects the degree of asymmetry of the airflow, and the greater the absolute value, the more serious the turbulence of the airflow caused by vortex or lateral leakage flow, and the higher the risk of UAV attitude disturbance; the sinking weight is a pre-set adjustment parameter for controlling the influence of the vortex asymmetry coefficient on the sinking speed (the greater the weight, the more obvious the speed reduction under the same coefficient), which can be adjusted according to the actual scene such as the UAV load and hangar environment. When the absolute value of the vortex asymmetry coefficient is small (the airflow is close to symmetry), the product of β and the absolute value is small, and 1 minus the product is close to 1, so the sinking speed is close to the initial sinking speed, ensuring the recovery efficiency; when the absolute value of the vortex asymmetry coefficient is large (the airflow is seriously asymmetrical), the product of β and the absolute value increases, and the value of 1 minus the product decreases, and the sinking speed decreases accordingly. The core purpose of this adjustment is that the more turbulent the airflow, the slower the UAV descends, thereby reserving more sufficient response time for the difference PID control (adjusting the duct motor speed), reducing the attitude deviation caused by rapid sinking, that is, when sinking slowly, even if there is airflow disturbance, the rate of attitude change is more gentle, and the control system has enough time to offset the disturbance by adjusting the motor speed, avoiding attitude instability.
[0063] The above describes the embodiments of the present embodiment, but the present embodiment is not limited to the specific embodiments described above, and the specific embodiments described above are only illustrative and not limiting. Those skilled in the art can make many forms under the inspiration of the present embodiment, which are all within the protection scope of the present embodiment.
Claims
1. A drone recovery control system based on a ducted motor, characterized in that, include: Critical determination unit, used to determine critical orientation difference based on the length and width of the target hangar; The image acquisition unit is used to acquire orthophotos at a fixed sampling frequency through a downward-looking camera, so as to extract the aerodynamic reflection zone based on the orthophotos; The fitting unit is used to perform straight line fitting based on the aerodynamic reflection zone to obtain a first straight line and a second straight line, and to determine the orientation angle, orientation angle difference and orientation angle mean of the first straight line and the second straight line in sequence. A symmetry processing unit is used to determine the vortex asymmetry coefficient based on the orientation angle difference and the critical orientation difference. The frequency processing unit is used to extract the mean values of M orientation angles within a preset time window and process them through Fourier transform to obtain the oscillation frequency of the reflection zone and the jitter frequency of the UAV. The recovery control unit is used to perform differential PID recovery control management on the drone based on the oscillation frequency of the reflector zone, the drone jitter frequency, and the vortex asymmetry coefficient, so as to achieve soft landing control of the drone.
2. The UAV recovery control system based on a ducted motor according to claim 1, characterized in that, Based on the length and width of the target hangar, the critical orientation difference is determined, including: Read the length and width of the target hangar opening; The arctangent of the width-to-length ratio is used as the critical orientation difference.
3. The UAV recovery control system based on a ducted motor according to claim 2, characterized in that, Extracting aerodynamic reflection zones based on orthophotos, including: The raw image is acquired using a downward-facing camera; The original image is transformed using a preset inverse perspective transformation matrix to form an orthophoto; Extract the orthophotos of the current time within a preset preceding time period and calculate the mean image; Calculate the difference between the orthophoto and the mean image at the current moment to form a difference image; The difference image is converted to grayscale and a grayscale threshold is set to form a first mask for binary classification; wherein, the first mask is a binary matrix, where element 1 represents a candidate reflective element and the rest are 0; A morphological closure process is performed on the first mask to form an initial aerodynamic reflection zone; Using the centroid of the initial aerodynamic reflection zone as the origin of the polar coordinates, K polar rays are generated; where the polar angle range of the polar rays is... The polar angle interval between adjacent polar rays is ; For the i-th polar ray, sample the initial aerodynamic reflection region outward from its centroid. If the sampled value changes from 1 to 0, extract the Euclidean distance from that element to its centroid. ; For the i-th polar ray, the rhythm value is calculated as follows: in, This represents the Euclidean distance of the (i+1)th polar ray. The Euclidean distance of the (i-1)th polar ray; If at least one exists If the rhythm value of the polarimetric rays is greater than the preset rhythm threshold, then the initial aerodynamic reflection zone is marked as valid; otherwise, return and adjust the grayscale threshold until at least one polarimetric ray is present. The aerodynamic reflection zone is obtained when the rhythm value of the polar rays exceeds the preset rhythm threshold.
4. The UAV recovery control system based on a ducted motor according to claim 3, characterized in that, Linear fitting is performed based on the aerodynamic reflection zone to obtain a first straight line and a second straight line. The orientation angle, the difference in orientation angle, and the mean orientation angle of the first and second straight lines are then determined sequentially, including: The boundary pixels of the aerodynamic reflection zone are sequentially extracted from the orthophoto to obtain the outer contour. Using the i-th pixel in the outer contour as the center, extract a preset number of pixels in the outer contour before and after it to calculate the curvature value of the i-th pixel. Extract the pixel with the largest curvature value from the outer contour as the segmentation pixel; The outer contour is divided into a first set of pixels and a second set of pixels by segmentation pixels; The first straight line and the second straight line are obtained by fitting the first pixel set and the second pixel set, respectively. The corresponding orientation angles are obtained by taking the arctangent of the slopes of the first and second lines, respectively. The difference between the orientation angles of the first and second lines is obtained by subtracting the orientation angle difference value. Calculate the average of the orientation angles of the first and second lines to obtain the mean orientation angle.
5. A drone recovery control system based on a ducted motor according to claim 4, characterized in that, Based on the orientation angle difference and critical orientation difference, the vortex asymmetry coefficient is calculated, including: The angle residual is obtained by subtracting the orientation angle difference from the critical orientation difference. The ratio of the angular residual to the preset normalized residual is used as the vortex asymmetry coefficient.
6. The UAV recovery control system based on a ducted motor according to claim 5, characterized in that, The mean values of M orientation angles within a preset time window are extracted and processed using Fourier transform to obtain the oscillation frequency of the reflector zone and the jitter frequency of the UAV, including: Extract the mean values of M orientation angles within a preset time window to form a sequence of mean orientation angles; Calculate the average value of the orientation angle mean sequence to obtain the DC component; The mean value of the orientation angle sequence is reduced by the DC component to obtain a zero-mean sequence. Construct a Hanning window function hn of length M, n∈ Where n is 1 or M, the Hanning window function takes the value of 0; when n is 1 or M, the value of n is M. At that time, the Hanning window function reaches its maximum value; The zero-mean sequence is weighted sequentially using the Hanning window function to obtain a preprocessed sequence; Determine the amplitude spectrum for the m-th unit in the preprocessed sequence, including: Perform Fourier transform on the m-th unit in the preprocessed sequence to obtain the complex spectrum, and use the modulus of the complex spectrum as the amplitude spectrum of the m-th unit; The product of the ratio of the fixed sampling frequency and M and m is used as the frequency of the m-th unit; If the amplitude spectrum of the m-th unit is the largest, then the frequency of the m-th unit is taken as the oscillation frequency of the reflection zone, and half of the oscillation frequency of the reflection zone is taken as the jitter frequency of the UAV according to the prior proportional relationship.
7. A drone recovery control system based on a ducted motor according to claim 6, characterized in that, Differential PID recovery control management for drones includes: The initial differential error is calculated based on the drone's jitter frequency, and then weighted by a vortex asymmetry coefficient to form the differential error. ; Determine the system's natural frequency Preset damping ratio ζ and total moment of inertia ;in, This represents the drone's jitter frequency and calculates the weight of the proportional term in the PID gain. Weight of integral terms and differential term weights ;in, , and ; Based on differential error Combined with the weight of the proportional term Weight of integral terms The proportional term error and integral term error of the PID gain are calculated using the differential term weights, respectively. and differential term error ; in, Indicates time The integral term error, Indicates time The differential error, Indicates a fixed sampling frequency; Differential speed commands are synthesized based on proportional term error, integral term error, and differential term error.
8. A drone recovery control system based on a ducted motor according to claim 7, characterized in that, Differential PID recovery control management for drones also includes: Determine the reference speeds of the left and right ducted motors of the target UAV. ; Reference speed The sum of the differential speed command and the target speed of the left bypass motor is used as the target speed. Reference speed The difference between the speed command and the differential speed command is used as the target speed of the right duct motor.
9. A drone recovery control system based on a ducted motor according to claim 8, characterized in that, Differential PID recovery control management for drones also includes: The descent speed of the target UAV is controlled based on the vortex asymmetry coefficient. ,as follows: in, This indicates the initial descent velocity of the target drone. This represents the vortex asymmetry coefficient. This indicates the weighting of the sinking.