WO2023130691A1

WO2023130691A1 - Dynamic gliding method and system based on distributed pressure sensors and segmented attitude control

Info

Publication number: WO2023130691A1
Application number: PCT/CN2022/103500
Authority: WO
Inventors: 谢芳芳; 季廷炜; 王丹翔; 陆宇峰; 郑耀
Original assignee: 浙江大学
Priority date: 2022-01-06
Filing date: 2022-07-02
Publication date: 2023-07-13
Also published as: CN114355777A; CN114355777B

Abstract

A dynamic gliding method and system based on distributed pressure sensors and segmented attitude control. The method comprises: acquiring the current state parameters of a glider in real time; by using a trained neural network, obtaining a nonlinear relationship between a surface pressure and a flow working condition, obtaining the current predicted value of a Reynolds number and the current predicted value of the angle of attack of the glider, and then obtaining an estimated value of a wind speed gradient; solving a target attitude of each phase in real time by performing derivation by means of a motion equation, by using a path chasing algorithm, i.e. carrot-chasing, and in combination with the estimated value of the wind speed gradient; and finally, controlling the current attitude of the glider to get close to a target attitude, so as to complete segmented attitude control. By means of the dynamic gliding segmented attitude control system, an unmanned aerial vehicle acquiring energy from wind shear is stimulated. The method is easily implemented and can be widely applied to the utilization of wind power of an unmanned aerial vehicle during wind shearing. The precision of estimation over wind is high, and an unmanned aerial vehicle can obtain sufficient energy to keep flying.

Description

A dynamic gliding method and system based on distributed pressure sensors and segmented attitude control

technical field

The invention relates to a dynamic gliding method and system based on distributed pressure sensors and segmented attitude control. In the simulation environment, the wind information at the current position can be estimated in real time more accurately, and the unmanned aerial vehicle can obtain sufficient wind information. Energy to maintain flight belongs to the field of wind field information perception and wind energy utilization on UAVs.

Background technique

Due to its low cost and convenient transportation and operation, small UAVs are widely used in areas such as area search, environmental monitoring, agricultural plant protection, and aerial photography. However, long-duration endurance for small drones remains a challenge due to limited battery capacity. Compared with rotary-wing UAVs, fixed-wing UAVs have higher energy efficiency and longer endurance, and have the potential to improve performance, such as using thin-film solar cells to absorb solar energy or adjusting attitude to harness wind energy. Albatrosses, who are good at flying in nature, use wind energy by adjusting their attitude up and down through the wind shear layer. This dynamic gliding action can be applied to the design of long-endurance drones.

Among them, a key issue in realizing dynamic gliding is the acquisition of real-time wind field information. At present, the wind speed is estimated mainly through the inertial sensor (measurement of ground speed) and pitot tube and wind vane (measurement of air speed), but the conventional pitot tube and wind vane It will affect the aerodynamic shape of the UAV and affect the aerodynamic performance, and only the average value can be obtained, and the dynamic (pulsating wind) information of the local flow cannot be captured, and the dynamic gliding action is easily affected by the aerodynamic performance and pulsating wind. Speed sensors may degrade the actual performance of dynamic glides. Another key issue to realize dynamic gliding is the control of dynamic gliding action. At present, the method of trajectory optimization and path following control is mainly used to simulate dynamic gliding action. This method regards the control of dynamic gliding as a nonlinear programming problem. Optimize the entire trajectory in one cycle, and control the UAV to follow the optimal trajectory optimized before. However, although this method can fly a similar trajectory, the actual capacitation will be obvious due to incorrect attitude. The problem of being less than the expected value may make the benefits of dynamic gliding actions not as good as the conventional cruising mode.

Contents of the invention

The purpose of the present invention is to address the limitations of the existing airspeed sensor that can affect the aerodynamic shape of the UAV and cannot capture dynamic flow information, and the shortcomings of the dynamic gliding method through trajectory optimization and path following control. Compared with the albatross dynamic gliding method and system of segmented attitude control, compared with the traditional method, the present invention has a higher estimation accuracy of airspeed, and will not affect the aerodynamic shape of the UAV, so that the UAV can obtain enough energy To keep flying, there will be no obvious shortage of energy gain.

The purpose of the present invention is achieved through the following technical solutions:

A dynamic gliding method based on distributed pressure sensors and segmented attitude control, the dynamic gliding includes P1: headwind climbing stage, P2: high-altitude turning transition stage, P3: downwind diving stage, P4: low-altitude turning transition stage, P5 : Level flight stage along the target direction; the specific method is:

Real-time acquisition of the current state parameters of the glider, including: the current coordinates of the glider, height h, heading angle ψ, pitch angle θ, roll angle μ, pressure data of multiple distributed pressure sensors arranged on the surface of the glider wing, and inertial sensor measurement The ground speed V _d and the side slip angle β measured by the side slip angle sensor.

Input the acquired pressure data (average value within a certain period of time) into a trained neural network model to obtain the predicted Reynolds number Re' and the predicted angle of attack α' of the current glider; and calculate according to the predicted Reynolds number Re' Obtain the predicted value of airspeed V′ _a ; combine the predicted value of airspeed V′ _a , ground speed V _d , sideslip angle β and real-time height h to calculate and obtain the estimated value β′ _w of the wind speed gradient of the x _i direction component in the ground coordinate system :

β′ _w =((W _ix ′) _t -(W _ix ′) _t-1 )/(h _t -h _t-1 )

W _ix '=V _d -V' _a cosα'cosβ

Wherein, the subscript t is the current moment, the subscript t-1 is the previous moment, and W _ix ′ is the wind speed of the x _i direction component.

Through the derivation of the motion equation and the carrot-chasing path following algorithm combined with the estimated value of the wind speed gradient _β′w , the target attitude of each stage is solved in real time. The target attitude includes heading angle, pitch angle, roll angle and sideslip angle, where:

Stage P1 and stage P3 are steady climbing or steady diving, target roll angle μ _d1 = 0 and μ _d3 = 0, target heading angle ψ _d1 = π-ψ _d3 , target pitch angle θ _d1 of stage P1 and target of stage P3 Pitch angle _θd3 is obtained by maximizing the gradient of energy with respect to altitude

The climb angle γ of the optimized track is obtained, and the remaining target attitudes are the set values:

where, e is the Oswald factor, S is the reference area of the wing, ρ is the density of the air, CD0 is the zero-lift drag coefficient, AR is the aspect ratio of the aircraft, S _{( )} and C _{( )} represent sin( ) and cos(·), γ is the climb angle of the track, χ is the azimuth of the track, g is the acceleration of gravity, and m is the mass of the glider.

The target roll angles μ _d2 and μ _d4 for phase P2 and phase P4 are maximized by maximizing the turning efficiency

To optimize, the remaining target poses are the set values:

The target heading angle ψ _d5 in stage P5 adopts the carrot-chasing two-dimensional path following algorithm, and is derived from the geometric relationship between the current coordinates of the glider and the path to be followed, and the remaining target attitudes are set values.

The current attitude of the glider is controlled to be close to the target attitude, and the segmented attitude control is completed.

Further, the distance between the plurality of distributed pressure sensors arranged on the surface of the wing of the glider is greater than 0.06c, where c is the average chord length of the wing.

Further, the positions of the multiple distributed pressure sensors arranged on the surface of the glider wing are determined by the following method:

Through the CFD numerical simulation of the two-dimensional flow field around the wing of the glider under different flow conditions, the pressure data of multiple coordinate points distributed on the surface of the wing corresponding to the different flow conditions are extracted to form the pressure data set X .

Using the measurement matrix C to represent the location information of the distributed pressure sensors, the relationship between the measurement data Y={y _i }, i=1,2,...,M and the data set X of the M distributed pressure sensors is expressed as Y=CX .

Dimensionality reduction is performed on the pressure data set X by the POD algorithm, and then by the sensor position and modality information

Items undergo QR decomposition, where

is the POD modal matrix after dimensionality reduction, and the measurement matrix C _o that is dominant for the data set X is selected. The position information of the sensor contained in the measurement matrix C _o is the determined multiple distributed pressures arranged on the surface of the glider wing The location of the sensor.

Further, the neural network model is obtained by training as follows:

Through CFD numerical simulation of the two-dimensional flow field around the glider wing under different flow conditions, the time-varying pressure data of multiple distributed pressure sensors arranged on the surface of the glider wing corresponding to different flow conditions are extracted and obtained form the training data set.

Construct the neural network model, use the training data set, take the pressure data of multiple distributed pressure sensors arranged on the surface of the glider wing as input, the Reynolds number prediction value Re' and the angle of attack prediction value α' as output, until the output prediction The value and the true value loss function converge, and the training obtains the neural network model.

Further, the neural network model is a BP neural network or a convolutional neural network.

A dynamic gliding simulation system for a glider, comprising:

The data acquisition module is used to obtain the current state parameters of the glider in real time, including: the current coordinates of the glider, the height h, the heading angle ψ, the pitch angle θ, the roll angle μ, and the parameters of multiple distributed pressure sensors arranged on the surface of the glider wing The pressure data, the ground speed V _d measured by the inertial sensor, and the side slip angle β measured by the side slip angle sensor.

The trained neural network model is used to input the obtained pressure data (average value within a certain period of time), and output the predicted Reynolds number Re' and angle of attack predicted value α' of the current glider.

The wind speed gradient estimator is used to calculate and obtain the wind speed gradient of the x _i direction component in the ground coordinate system by combining the Reynolds number predicted value Re′, the airspeed predicted value V′ _a , the ground speed V _d , the sideslip angle β and the real-time height h Estimated value β′ _w .

The target attitude solver is used to solve the target attitude of each stage in real time through the derivation of the motion equation and the carrot-chasing path following algorithm combined with the estimated value β′ _w of the wind speed gradient. The target attitude includes heading angle, pitch angle, roll angle and sideslip angle, where:

To optimize, the remaining target poses are the set values:

The controller controls the current attitude of the glider to approach the target attitude.

Beneficial effects of the present invention:

1. Through the data-driven POD algorithm, the airfoil surface pressure data is reduced in dimension, and the position of the sparse sensor is selected through QR decomposition to obtain the optimal distributed pressure sensor position, and the sensor position can be selected without relying on experience.

2. The distributed pressure sensor is applied to the dynamic gliding process, and the BP neural network model trained offline is used to estimate the airspeed in real time, which can reduce the influence of the sensor on the aerodynamic shape of the UAV, and can capture the surrounding dynamic flow information , which is suitable for complex operations such as dynamic gliding of UAVs.

3. By dividing a period of dynamic gliding trajectory into several energy-acquiring segments and energy-consuming segments, and the PID controller group controls the current attitude of the glider to approach the calculated target attitude to achieve dynamic gliding operations, which can allow the UAV to obtain enough energy to maintain flight, and there will be no situation where dynamic gliding energy gain is obviously insufficient due to incorrect attitude.

Description of drawings

Fig. 1 is a flow chart of the present invention;

Figure 2 is the location of the optimized distributed pressure sensor and the corresponding pressure distribution diagram;

Figure 3 is a BP neural network structure diagram for estimating flow conditions;

Figure 4 is a schematic diagram of the track coordinate system;

Fig. 5 is the structural diagram of the dynamic gliding subsection attitude control system of unpowered glider;

Fig. 6 is a schematic diagram of dynamic gliding segmentation attitude control and stage switching logic;

Fig. 7 is a schematic diagram of a dynamic gliding trajectory of a cycle in the test example;

Fig. 8 is a diagram of energy changes in the simulated dynamic gliding process provided by the present invention;

Fig. 9 is a diagram of wind speed estimation results in the simulated dynamic gliding process provided by the present invention;

Fig. 10 is a graph of wind gradient estimation results in the simulated dynamic gliding process provided by the present invention.

Detailed ways

The present invention will be described in detail below in conjunction with specific embodiments according to the accompanying drawings.

The embodiment of the present invention adopts the glider of large aspect ratio as object, adopts straight wing, and airfoil is NACA 4412, and the relevant parameter of glider is as shown in table 1 below:

Table 1 Relevant parameters of the glider

Wherein, x _b , y _b , and z _b are three coordinate axes on the fuselage coordinate system.

The CFD numerical simulation example is the two-dimensional flow field around the NACA 4412 airfoil under the Reynolds number from 0.82e5 to 3.26e5 and the angle of attack from 0° to 22°, extracting 400 coordinate points uniformly distributed on the airfoil surface The pressure data, as the surface pressure distribution of the airfoil, provides a data set for the selection of the sensor position and the training of the neural network model.

Fig. 1 is a flow chart of a dynamic gliding simulation method based on distributed pressure sensors and segmented attitude control provided by the present invention. As shown in Fig. 1, the method proposed by the present invention specifically includes:

Offline training phase:

The CFD numerical simulation of the two-dimensional flow field around the NACA4412 airfoil under different flow conditions (0.82e5≤Re≤3.26e5, 0°≤α≤22°) was carried out through the Openfoam program, and the SA model in the RANS turbulence model was used to solve the problem For the unsteady turbulent flow field, the pressure data of 400 coordinate points uniformly distributed on the airfoil surface are extracted, and the interval of the time series is selected as Δt=1×10 ^-2 s, a total of 5000 time slices, and the time-dependent changes of different flow conditions are obtained. The pressure data set X, the pressure data set X is a two-dimensional matrix, wherein the number of rows represents the number of coordinate points, and the number of columns represents the number of time slices.

According to the data set X, the location of the distributed pressure sensor on the airfoil surface is selected: the measurement matrix C is used to represent the position information of the distributed pressure sensor, then the measurement data of M distributed pressure sensors Y={y _i },i =1, 2,..., the relationship between M and data set X can be expressed as Y=CX, where the minimum distance between adjacent distributed pressure sensors is 0.06c, and c is the average chord length of the wing, so as to keep Make enough room to install the airway at the pressure tap.

As a preferred solution, the data set X can be dimensionally reduced by the POD algorithm to obtain a low-dimensional representation of the data set X, that is, the POD modal matrix

and the corresponding POD coefficient matrix a _k , where k represents the number of modes; then by using the sensor position and mode information

Items are QR decomposed, and the measurement matrix C _o that is dominant for the data set X is selected;

Specifically include the following sub-steps:

1) Singular value decomposition is performed on the obtained pressure data set X to obtain the left singular matrix

And the right singular matrix V, and the characteristic matrix Σ with eigenvalues (singular values), and only keep the first k according to the singular values to reduce the dimensionality of the data set X:

in,

is the POD modal matrix corresponding to the first k singular values, and a _k is the corresponding POD coefficient matrix, which is given by the orthogonal projection:

2) When the data set X as the state is unknown and the measurement data Y is known, the POD coefficient is approximated by the Moore-Penrose pseudo-inverse

and rebuild the state

Among them, Θ ⁺ is the pseudo-inverse of matrix Θ, and M is the number of sensors.

3) The optimal pressure sensor position C _o is to reconstruct the state

The position closest to the real state X, then optimize the inverse of Θ to obtain the position C _o of the pressure sensor:

C _o ＝argmax|det(M _C )|,M _C ＝Θ ^T Θ

det(M _C ) represents the determinant of the computation matrix M _C .

The positions of the distributed pressure sensors optimized in this embodiment and the corresponding pressure distribution are shown in Figure 2. The number of sensors obtained in this embodiment is 11, which are mainly distributed on the upper surface of the airfoil, and the distance between adjacent sensors is greater than or equal to The minimum spacing is 0.06c.

Construct the neural network model for estimating the flow conditions, the neural network model can be a conventional neural network model, the present embodiment is a BP neural network model, and the structure of the BP neural network model is divided into three layers: input layer, hidden layer and output layer ,As shown in Figure 3. The input in the input layer is the distributed pressure sensor measurement value pressure data (the mean value that each sensor measures in a period of time, the mean value of all time slices is taken in this embodiment), and the hyperbolic tangent sigmoid function is used as activation in the hidden layer Function, the output in the output layer is the predicted value of Reynolds number Re and angle of attack α of flow conditions. The present embodiment is provided with two hidden layers, and the number of neurons of each layer is set (in this embodiment, the input layer is 11, the two hidden layers are respectively 11 and 11, and the output layer is 2), using trainbr The Bayesian regularization method is used as the backpropagation algorithm, and the optimizer learning rate is 0.001. Use the obtained input and output data to train the BP neural network model until the output prediction value and the true value loss function converge, so as to obtain the trained BP neural network model.

Online control stage:

The energy-capturing mechanism of dynamic gliding is deduced through the equation of motion, assuming that the wind velocity vector only considers the component W _ix in the direction of x _i , in this embodiment, by

To simulate the wind shear layer, the reference velocity W ₀ =10m/s, the reference height z ₀ =10m, the thickness parameter l=1, h is the real-time height of the unpowered glider, in the track coordinate system (as shown in Figure 4 ) to obtain the motion equation of the six degrees of freedom of the unpowered glider:

Among them, x _i y _i z _i represents the ground coordinate system, S _{( )} and C _{( )} represent sin ( ) and cos ( ), L and D are lift and drag, V _a is airspeed, χ is airspeed where m is the mass of the unpowered glider, g is the gravitational acceleration, h is the real-time height of the unpowered glider, γ is the climb angle of the track, and μ is the aerodynamic roll angle;

Respectively represent V _a , χ, γ, and the time derivative of the displacement of the glider in the directions of xi _, y _{i and} _zi . The total energy of the glider in the track coordinate system is defined as

The total energy E is derived with respect to time t:

Among them, -DV _a is the item of resistance consumption, and the wind gradient β _w = -dW _ix /dz _i is the key to dynamic gliding energy capture. To obtain energy for dynamic gliding, then -mV _a ² β _w S _γ C _γ C _χ must be is greater than 0, and the wind gradient β _w >0, then S _γ C _γ C _χ must be less than 0. It is concluded that when climbing (0<γ<π/2), the glider needs to headwind (π/2<χ<3π /2) to gain energy; when diving (-π/2<γ<0), the glider needs to be downwind (-π/2<χ<π/2) to gain energy.

Therefore, the present invention is based on the dynamic gliding method of distributed pressure sensors and segmented attitude control, that is, the online control stage is specifically:

Input the obtained pressure data (averaged in a fixed time) into the trained neural network model to obtain the predicted Reynolds number Re' and the predicted angle of attack α' of the current glider; Predicted airspeed value V′ _a ; combined with predicted airspeed value V′ _a , ground speed V _d , side slip angle β and real-time height h, the estimated value β′ _w of the wind speed gradient of the _xi direction component in the ground coordinate system is obtained.

Through the derivation of motion equations and carrot-chasing path-following algorithm combined with the estimated value of wind speed gradient _β′w , the target attitude of each stage is solved in real time. The target attitude includes heading angle, pitch angle, roll angle and sideslip angle.

Finally, the current attitude of the glider is controlled to be close to the target attitude, and the segmented attitude control is completed.

In the present embodiment, in order to describe the present invention in detail, in the Simulink program, the dynamic gliding segmentation attitude control system (as shown in Figure 5) of a unpowered glider has been built, and this system comprises wind velocity field simulator, six degrees of freedom unpowered Glider model, PID controller set, target attitude solver. Among them, the six-degree-of-freedom unpowered glider model is obtained from the Aerosim module. The input of the glider model is the wind speed field and the control surface commands (aileron, flap, elevator, rudder), and the output is the current state of the glider, including the real-time height h of the glider , speed, acceleration, airspeed, etc., and further obtain the current attitude of the glider: heading angle ψ, pitch angle θ, roll angle μ, sideslip angle β. The trajectory of a dynamic gliding cycle is divided into 5 segments: P1 upwind climb, P2 high-altitude turn transition, P3 downwind dive, P4 low-altitude turn transition, P5 level flight along the target direction, UAV climbs against the wind (P1) and under the downwind Energy is obtained from the wind field when diving (P3), and the extra energy is used for energy consumption in the turning transition process (P2, P3) and the peaceful flight process (P5). Among them, when h≥h _max , stage P1 switches to stage P2, when the heading angle |ψ-ψ _d3 |≤3°, stage P2 switches to stage P3, and when h≤h _min , stage P3 switches to stage P4 , when the total energy E>E ₀ and |ψ-ψ _d5 |≤6 _° , stage P4 _switches to stage P5; It is to enter the next P1 to obtain energy. Among them, h _max and h _min are the set maximum altitude and minimum altitude respectively, ψ _d1 is the target heading angle of stage P1, ψ _d3 is the target heading angle of stage P3, and ψ _d5 is the target heading angle of stage P5. E ₀ is the initial total energy. The target attitude of each segment is solved by the motion equation derivation and carrot-chasing path following algorithm, and then the current attitude of the glider is controlled to approach the target attitude through the PID controller group, as shown in Figure 6.

Among them, the target attitude of each segment is solved through motion equation derivation and carrot-chasing path following algorithm. The specific process is as follows:

Obtain the pressure data y′ _i (i=1,2,…,M) of the glider at the current moment, and input it into the trained BP neural network model, the Reynolds number prediction value Re′ and the angle of attack prediction value α′, and further obtain The space velocity V′ _a under this flow condition,

μ ₀ is the viscosity coefficient of the air, and l ₀ is the characteristic length, which is generally expressed by the average chord length c of the wing on an airplane. The estimated airspeed is then combined with the ground speed _Vd measured by the inertial sensor and the measured value β of the sideslip angle sensor. Get an estimate of the wind speed:

W _ix '=V _d -V' _a cosα'cosβ

The wind speed V _wx ′ is numerically differentiated together with the real-time height h of the glider to obtain the estimated wind gradient β′ _w :

β′ _w =dW _ix /dz _i ′=((W _ix ′) _t -(W _ix ′) _t-1 )/(h _t -h _t-1 )

Wherein, the subscript t is the current moment, and the subscript t-1 represents the previous moment.

According to the estimated wind gradient _β′w , optimize and solve the target attitude of each segment, where:

Assuming target heading angle ψ _d1 = π-ψ _d3 in phases P1 and P3, target pitch angle θ _d1 = γ _opt1 for phase P1 and target pitch angle θ _d3 = γ _opt3 for phase P3 by maximizing the gradient of energy with respect to altitude

to optimize, and the remaining target postures are set values based on experience (as shown in Figure 6):

γ _opt1 represents the track climb angle obtained by maximizing the gradient optimization of energy relative to height in stage P1, γ _opt3 represents the path climb angle obtained by maximizing the gradient optimization of energy relative to height in stage P3, e is the Oswald factor, S is the reference area of the wing, ρ is the density of the air, CD0 is the zero-lift drag coefficient, AR is the aspect ratio of the aircraft, and it is assumed that the UAV climbs steadily or dives stably (target μ=0, dγ/dt=0 ; target χ=0 or π).

Among them, the target roll angles μ _d2 and μ _d4 in phases P2 and P4 are maximized by maximizing the turning efficiency

To optimize, the rest of the target postures are set values based on experience (as shown in Figure 6):

Among them, S _γ is a function of μ. Under the constraints of the maximum lift coefficient CL _max and the maximum load coefficient n _max , the fitting curve obtains the optimal roll angle related to the airspeed. In this embodiment, the fitting curve is

Among them, the heading angle ψ _d5 in stage P5 is derived from the geometric relationship between the current coordinates of the UAV and the path to be followed by using the carrot-chasing two-dimensional path following algorithm, and the rest of the target attitudes are set values based on experience.

This embodiment uses the current airspeed and angle of attack state of the glider and the obtained pressure data under different airspeed and angle of attack conditions to interpolate the current pressure data of the glider as the pressure data at the current moment of the glider for online simulation. The test results are as follows As shown in Figures 7 to 10, in the test, the unpowered glider performed a cycle of dynamic gliding action in a wind shear environment. The horizontal wind direction is along the negative direction of the x-axis. , yaw angle ψ _d1 =0, pitch angle θ=0, roll angle μ=0 and start to enter the dynamic gliding action, and the motion track is shown in Fig. 7 . According to the test results, the unpowered glider can obtain enough energy from the wind field when climbing against the wind (P1) and diving with the wind (P3), and the excess energy is used in turning (P2, P4) and flying peacefully (P5). consumed during the process, as shown in Figure 8. According to the wind speed estimation results shown in Figure 9, it can be seen that the estimated wind speed profile is basically consistent with the real wind speed profile in most of the time, and there will be a small amount of error when switching stages, but the overall root mean square error of the wind speed remains at 0.5m Within /s. According to the estimation results of the wind gradient, compared with the real data, the numerical differentiation method used in the estimation will bring a time delay of 0.22s. Figure 10 shows the estimation of the wind gradient after removing the time delay. When climbing against the wind and diving with the wind (the height is 5m to 15m), the key wind gradient information obtained through dynamic gliding will change drastically. It can be seen from Figure 10 that the estimated wind gradient outline at this time is different from the real wind gradient The contours are basically matched, giving accurate wind gradient information.

Corresponding to the aforementioned embodiments of the dynamic gliding method based on distributed pressure sensors and segmented attitude control, the present invention also provides an embodiment of a dynamic gliding simulation system for a glider.

A dynamic gliding simulation system for a glider provided by an embodiment of the present invention includes one or more processors for implementing the dynamic gliding method based on distributed pressure sensors and segmented attitude control in the above embodiments.

The embodiment of the glider dynamic gliding simulation system of the present invention can be applied to any device with data processing capability, and any device with data processing capability can be a device or device such as a computer.

Exemplarily, a dynamic gliding simulation system for a glider includes:

To optimize, the remaining target poses are the set values:

Apparently, the above-mentioned embodiments are only examples for clear description, rather than limiting the implementation. For those of ordinary skill in the art, other changes or changes in different forms can be made on the basis of the above description. It is not necessary and impossible to exhaustively list all implementation modes here. However, the obvious changes or variations derived therefrom still fall within the protection scope of the present invention.

Claims

A dynamic gliding method based on distributed pressure sensors and segmented attitude control, the dynamic gliding includes P1: headwind climbing stage, P2: high-altitude turning transition stage, P3: downwind diving stage, P4: low-altitude turning transition stage, P5 : level flight stage along the target direction; it is characterized in that the method is specifically:

Real-time acquisition of the current state parameters of the glider, including: the current coordinates of the glider, height h, heading angle ψ, pitch angle θ, roll angle μ, pressure data of multiple distributed pressure sensors arranged on the surface of the glider wing, and inertial sensor measurement The ground speed V d and the side slip angle β measured by the side slip angle sensor.

Input the obtained pressure data into a trained neural network model to obtain the predicted Reynolds number Re' and the predicted angle of attack α' of the current glider; and calculate the predicted airspeed V'a based on the predicted Reynolds number Re'; Combined with the airspeed prediction value V′ a , ground speed V d , sideslip angle β and real-time height h, the estimated value β′ w of the wind speed gradient of the x i direction component in the ground coordinate system is obtained:

β′ w =((W ix ′) t -(W ix ′) t-1 )/(h t -h t-1 )

W ix '=V d -V' a cosα'cosβ

Wherein, the subscript t is the current moment, the subscript t-1 is the previous moment, and W ix ′ is the wind speed of the x i direction component.

Through the derivation of the motion equation and the carrot-chasing path following algorithm combined with the estimated value of the wind speed gradient β′w , the target attitude of each stage is solved in real time. The target attitude includes heading angle, pitch angle, roll angle and sideslip angle, where:

Stage P1 and stage P3 are steady climbing or steady diving, target roll angle μ d1 = 0 and μ d3 = 0, target heading angle ψ d1 = π-ψ d3 , target pitch angle θ d1 of stage P1 and target of stage P3 Pitch angle θd3 is obtained by maximizing the gradient of energy with respect to altitude
The climb angle γ of the optimized track is obtained, and the remaining target attitudes are the set values:

where, e is the Oswald factor, S is the reference area of the wing, ρ is the density of the air, CD0 is the zero-lift drag coefficient, AR is the aspect ratio of the aircraft, S ( ) and C ( ) represent sin( ) and cos(·), γ is the climb angle of the track, χ is the azimuth of the track, g is the acceleration of gravity, and m is the mass of the glider.

The target roll angles μ d2 and μ d4 for phase P2 and phase P4 are maximized by maximizing the turning efficiency
To optimize, the remaining target poses are the set values:

The target heading angle ψ d5 in stage P5 adopts the carrot-chasing two-dimensional path following algorithm, and is derived from the geometric relationship between the current coordinates of the glider and the path to be followed, and the remaining target attitudes are set values.

The current attitude of the glider is controlled to be close to the target attitude, and the segmented attitude control is completed.
The dynamic gliding method according to claim 1, characterized in that the distance between the plurality of distributed pressure sensors arranged on the surface of the wing of the glider is greater than 0.06c, where c is the average chord length of the wing.
The dynamic gliding method according to claim 1, wherein the positions of a plurality of distributed pressure sensors arranged on the wing surface of the glider are determined by the following method:

Through the CFD numerical simulation of the two-dimensional flow field around the wing of the glider under different flow conditions, the pressure data of multiple coordinate points distributed on the surface of the wing corresponding to the different flow conditions are extracted to form the pressure data set X .

Using the measurement matrix C to represent the location information of the distributed pressure sensors, the relationship between the measurement data Y={y i }, i=1,2,...,M and the data set X of the M distributed pressure sensors is expressed as Y=CX .

Dimensionality reduction is performed on the pressure data set X by the POD algorithm, and then by the sensor position and modality information
Items undergo QR decomposition, where
is the POD modal matrix after dimensionality reduction, and the measurement matrix C o that is dominant for the data set X is selected. The position information of the sensor contained in the measurement matrix C o is the determined multiple distributed pressures arranged on the surface of the glider wing The location of the sensor.
The dynamic gliding method according to claim 1, wherein the neural network model is obtained by training as follows:

Through CFD numerical simulation of the two-dimensional flow field around the glider wing under different flow conditions, the time-varying pressure data of multiple distributed pressure sensors arranged on the surface of the glider wing corresponding to different flow conditions are extracted and obtained form the training data set.

Construct the neural network model, use the training data set, take the pressure data of multiple distributed pressure sensors arranged on the surface of the glider wing as input, the Reynolds number prediction value Re' and the angle of attack prediction value α' as output, until the output prediction The value and the true value loss function converge, and the training obtains the neural network model.
The dynamic gliding method according to claim 1, wherein the neural network model is a BP neural network or a convolutional neural network.
A dynamic gliding simulation system for a glider, characterized in that it comprises:

The data acquisition module is used to obtain the current state parameters of the glider in real time, including: the current coordinates of the glider, the height h, the heading angle ψ, the pitch angle θ, the roll angle μ, and the parameters of multiple distributed pressure sensors arranged on the surface of the glider wing The pressure data, the ground speed V d measured by the inertial sensor, and the side slip angle β measured by the side slip angle sensor.

The trained neural network model is used to input the obtained pressure data, and output the predicted Reynolds number Re' and the predicted angle of attack α' of the current glider.

The wind speed gradient estimator is used to calculate and obtain the wind speed gradient of the x i direction component in the ground coordinate system by combining the Reynolds number predicted value Re′, the airspeed predicted value V′ a , the ground speed V d , the sideslip angle β and the real-time height h Estimated value β′ w .

The target attitude solver is used to solve the target attitude of each stage in real time through the derivation of the motion equation and the carrot-chasing path following algorithm combined with the estimated value β′ w of the wind speed gradient. The target attitude includes heading angle, pitch angle, roll angle and sideslip angle, where:

Stage P1 and stage P3 are steady climbing or steady diving, target roll angle μ d1 = 0 and μ d3 = 0, target heading angle ψ d1 = π-ψ d3 , target pitch angle θ d1 of stage P1 and target of stage P3 Pitch angle θd3 is obtained by maximizing the gradient of energy with respect to altitude
The climb angle γ of the optimized track is obtained, and the remaining target attitudes are the set values:

where, e is the Oswald factor, S is the reference area of the wing, ρ is the density of the air, CD0 is the zero-lift drag coefficient, AR is the aspect ratio of the aircraft, S ( ) and C ( ) represent sin( ) and cos(·), γ is the climb angle of the track, χ is the azimuth of the track, g is the acceleration of gravity, and m is the mass of the glider.

The target roll angles μ d2 and μ d4 for phase P2 and phase P4 are maximized by maximizing the turning efficiency
To optimize, the remaining target poses are the set values:

The target heading angle ψ d5 in stage P5 adopts the carrot-chasing two-dimensional path following algorithm, and is derived from the geometric relationship between the current coordinates of the glider and the path to be followed, and the remaining target attitudes are set values.

The controller controls the current attitude of the glider to approach the target attitude.