Professional Documents
Culture Documents
Presence of Wind
Aseem S. Nevrekar1, Alfred G. Striz2, and Prakash Vedula3
School of Aerospace and Mechanical Engineering, The University of Oklahoma, Norman OK
73019-1052
supersonic interceptor aircraft is presented in this paper, and the influence of wind
dimensional motion of a point mass model of the aircraft. Comparisons are made
between cases without wind, an actual wind profile varying with altitude, and a
trajectories for the three cases are obtained for different initial heading angles
covering the first quadrant, along with the corresponding state and control variable
histories. Most of the results for the variable wind case and the corresponding
average wind case are found to be close. This observation suggests that the
assumption of a constant wind profile with the correct average wind velocity,
obtained from the actual variable wind profile, will lead to reasonably accurate
predictions of optimal glide trajectories for maximum range, as well as for the
1
Graduate Student
2
Professor Emeritus
3
Associate Professor
1
Nomenclature
𝐶𝐷 = drag coefficient
𝐶𝐿 = lift coefficient
𝐷 = drag
ℎ = altitude
𝐽 = objective function
𝐿 = lift
𝑀 = Mach number
𝑚 = mass of aircraft
𝑞 = dynamic pressure
𝑅 = range
𝑊 = weight of aircraft
2
𝑥𝑓 = distance travelled in x-direction along ground plane
𝛼 = angle of attack
𝜌 = atmospheric density
𝜙 = bank angle
𝜒 = heading angle
I. Introduction
Extensive research has been conducted in the field of trajectory optimization since the 1950s
due to the availability of digital computers. Early work on an optimal control approach applied to
trajectory optimization was conducted by Bliss1, Pontryagin2, Bryson3 and Bellman4. Bellman
generalized the classical Hamilton-Jacobi theory to obtain optimal control functions and optimal
objectives for all possible initial conditions (based on the solution of a linear partial differential
equation). In the early 1960s, a simple Newton method was commonly used for trajectory
optimization, and the optimization was implemented parametrically5. Later on, in the 1970s, a
reduced gradient method6-8 began to be used for solving constrained optimization problems, for
3
nearly linear constraints and a small number of inequalities. One of the first applications of
trajectory optimization techniques was to optimize rocket thrust profiles in vacuum and in the
atmosphere. Bryson and Denham9,10 (1962) used a steepest-ascent method for solving optimum
programming problems. Bryson, Desai and Hoffman11 (1969) used the energy-state
In more recent years, pseudospectral methods have increasingly been used to solve trajectory
optimization problems owing to their high degree of accuracy. Fahroo and Ross12 demonstrated
considering the canonical Brachistochrone problem and the lunar landing problem. Dai and
Cochran13 applied CPM and the Gauss Pseudospectral Method to solve two-dimensional and
three-dimensional MTTC and MFTC problems for aircraft. Dekel and Ben-Asher14 used
pseudospectral techniques to solve optimal glide problems involving low altitude, low velocity
flight conditions.
Optimal glide problems are important, especially in cases involving instantaneous engine
failure. When the engine of an aircraft in flight is powered off, it becomes necessary to determine
the maximum range the aircraft can cover in this condition - subject to existing constraints - and
the corresponding trajectories so that the feasibility of an emergency landing within this distance
can be determined. It is reasonable to expect that the presence of wind affects the maximum
range travelled by the aircraft and the underlying optimal trajectory, in glide condition.
This paper studies the effects that different wind profiles have on the optimum glide trajectory
and on the state and control variables. CPM is employed to transform the maximum range
problem from its optimal control form to a Non-Linear Programming (NLP) problem. CPM has
4
an advantage over other collocation methods due to the high degree of accuracy attainable. This
is because the Chebyshev-Gauss-Lobatto (CGL) node points used in CPM result in interpolating
polynomials that are closest to the optimal polynomial in the max-norm approximation of a
function12,15. The derivatives of these interpolating polynomials are calculated at the CGL node
points exactly, using a differentiation matrix. CPM thus offers relatively rapid solutions to
trajectory optimization problems. The resulting NLP problem can be solved using the NPSOL16
5.0 or SNOPT17,18 6.2 solvers. Both these solvers utilize a Sequential Quadratic Programming
The trajectory optimization problem can be considered as an Optimal Control Problem (OCP),
The control functions 𝑢(𝑡) are chosen to optimize the cost function19
𝐽 = 𝛷[𝒙(𝑡𝑓 ), 𝑡𝑓 ],
𝒈𝑙 ≤ 𝒈[𝒙(𝑡), 𝒖(𝑡), 𝒑, 𝑡] ≤ 𝒈𝑢 ,
5
the bounds on the state variables,
𝒙𝑙 ≤ 𝒙(𝑡) ≤ 𝒙𝑢 ,
𝒖𝑙 ≤ 𝒖(𝑡) ≤ 𝒖𝑢 ,
where 𝑡 is the independent variable (usually time), 𝒙 represents the vector of state variables, 𝒖 is
the vector of control variables and 𝒑 are the parameters not dependent on 𝑡. The subscripts "0"
and "𝑓" denote the initial and final conditions, respectively, while 𝑙 and 𝑢 denote the lower and
The Chebyshev Pseudospectral Method (CPM) discretizes the continuous solution of an OCP
represented by state and control variables by using orthogonal polynomials, which are infinitely
differentiable global functions, for interpolation to satisfy the differential equations, transforming
the OCP to a Non-Linear Programming Problem (NLPP). Examples of such polynomials are
Legendre and Chebyshev polynomials, which are orthogonal over the interval 𝝉 ∈ [−1, 1], with
respect to an appropriate weight function: 𝑤(𝜏) = 1 for Legendre polynomials, and 𝑤(𝜏) =
1⁄√1 − 𝜏 2 for Chebyshev polynomials of the first kind20. Since the exact solution to the OCP is
given by infinitely many state and control variable values, CPM is an approximation. It uses
Chebyshev-Gauss-Lobatto (CGL) collocation to locate the nodes required for discretization. The
𝜏𝑘 = cos(𝜋𝑘⁄𝑁), 𝑘 = 0, … , 𝑁 (1)
These nodes lie in the interval [-1,1] and are the extrema of the Nth order Chebyshev
6
𝑇𝑗 (𝜏) = cos(𝑗 cos −1 𝜏), 𝑗 = 0, … , 𝑁 (2)
which results in
This yields 𝑇0 (𝜏) = 1, 𝑇1 (𝜏) = 𝜏, 𝑇2 (𝜏) = 2𝜏 2 − 1, 𝑇3 (𝜏) = 4𝜏 3 − 3𝜏, and so on. These
nodes cluster around the end points of the interval. Interpolation at these nodes gives the results
where 𝑡0 is the initial time and 𝑡𝑓 the final time. The state variables 𝒙(𝝉) and control variables
𝒖(𝝉) can then be approximated using 𝑁𝑡ℎ order Lagrange interpolating polynomials as
𝒙𝑁 (𝝉) ≈ ∑𝑁
𝑗=0 𝒙𝑗 𝜙𝑗 (𝝉) (5)
𝒖𝑁 (𝝉) ≈ ∑𝑁
𝑗=0 𝒖𝑗 𝜙𝑗 (𝝉) (6)
where
𝝉−𝜏𝑗
𝜙𝑗 (𝝉) = ∏𝑁
𝑙=1 𝜏 (7)
−𝜏
𝑙≠𝑗 𝑗 𝑙
𝑘 = 0, … , 𝑁 (8)
7
(−1)𝑗+𝑘
−(𝑐𝑘 ⁄𝑐𝑗 ) [ (𝜏 ], 𝑗 ≠ 𝑘
𝑘 −𝜏𝑗 )
(2𝑁 2 +1)
− , 𝑗=𝑘=0
𝐷𝑘𝑗 = 𝐷𝑘𝑗 = 6 (9)
(2𝑁 2 +1)
− , 𝑗=𝑘=𝑁
6
𝜏𝑘
{ , 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
2(1−𝜏𝑘 2 )
𝑑 (𝑡𝑓 −𝑡0 )
𝒅𝑘 = 𝒙𝑁 (𝜏𝑘 ) − f(𝒙𝑘 , 𝒖𝑘 , 𝑡𝑘 ) = 0 𝑘 = 0, … , 𝑁 (10)
𝑑𝑡 2
The derivatives of the state variables are thus calculated (with a high degree of accuracy)
using the function values at the CGL nodes and the differentiation matrix 𝑫.
b. NLPP Formulation
After discretization using CPM, the above OCP is converted to a NLPP 19 to optimize the
𝒙𝑙 ≤ 𝒙 ≤ 𝒙𝑢
𝒃𝑙 ≤ 𝑨𝒙 ≤ 𝒃𝑢
𝒄𝑙 ≤ 𝒄(𝒙) ≤ 𝒄𝑢
where 𝒙 represents the vector of unknown state and control variable values and the final time 𝑡𝑓
(if 𝑡𝑓 is unknown). If 𝑁 is the number of discretization nodes (CGL points) and 𝐾 is the total
elements if 𝑡𝑓 is unknown.
8
III. Aircraft Dynamic Model
ℎ̇ = 𝑣 sin 𝛾 (11)
The parameters h, v, γ, χ, x and y are the state variables, and α and ϕ are the control variables.
Box/inequality constraints on the state and control variables, which also determine the upper and
0 ft ≤ ℎ ≤ 70,000 ft (17)
50𝜋 50𝜋
− 180 rad ≤ 𝜙 ≤ rad (22)
180
Assuming the aircraft is flying at an altitude of 44,820 ft at Mach 1.35 when the engine shuts
down instantaneously and the aircraft begins to glide, the initial conditions for true airspeed,
flight path angle and position are given by v0 = 1,307.04 ft/s, γ0 = 0 rad, and (x0 , y0, h0) = (0, 0,
44,820) ft, respectively. The final altitude is assumed to be zero, i.e., hf = 0 ft. The subscripts "0"
9
and "𝑓" are used to denote the initial and final conditions, respectively. The optimization
problem will involve finding the maximum value of the objective function R = (𝑥𝑓2 + 𝑦𝑓2 )1/2 with
The aircraft under consideration for glide optimization is a McDonnell Douglas F-4 Phantom
II supersonic interceptor, used for MTTC optimization in Reference 22. The aircraft weighs
The aerodynamic characteristics of the aircraft, including the effects of compressibility, are
shown in Figure 2. In particular, 𝐶𝐷0 , 𝐶𝐿𝛼 , 𝜂 and k are shown as a function of Mach number. A
piecewise cubic Hermite interpolation approach is used for the presentation of the data in Figure
1976 Standard Atmosphere23, are shown in Figures 3 and 4. These functional dependencies will
This allows for the calculation of lift and drag. Here, the parameters 𝐶𝐷0 , 𝐶𝐿𝛼 , and η are
obtained from
fromFigure
Figure2,2,and
andρ is obtained
ρ is from
obtained Figure
from 3. The
Figure lift and
3. The lift drag
and coefficients are calcula
drag coefficients are
calculated as follows:
11
CD = 𝐶𝐷0 + η𝐶𝐿𝛼 α2
CL = 𝐶𝐿𝛼 α
q = 0.5 ρ v2
L = q S CL
D = q S CD
The actual varying wind profile used in this paper is obtained from a nominal wind profile
defined by the centroids of the KSC 99% enveloping ellipse derived from monthly bivariate
An average velocity of this wind profile over altitude, is found to be 66.31 ft/s in the x-
direction and 6.62 ft/s in the y-directon, using the trapezoidal rule.
12
The values of wx and wy used in Equations (15) and (16) for both the actual varying and the
Using initial heading angle values of 𝜒0 = 0o, 30o, 60o and 90o, maximum range glide
trajectories along with the time histories of the state and control variables are obtained for cases
(i) with no wind, (ii) with the actual varying wind, and (iii) with its constant average
approximation. The solver NPSOL 5.0 is used for optimization16. The results presented in this
section were obtained using 50 CGL nodes. At this grid resolution, the results were found to be
converged.
The variation of altitude with horizontal distance is shown in Figure 6, for cases (i) - (iii),
when the initial heading angle is set to zero. When the engine is powered off, the aircraft uses its
speed for an initial increase in altitude with an associated decrease in velocity, which reduces
drag and increases potential energy, resulting in a larger glide range. As observed in the
13
magnified inset, the no wind and actual wind profiles cross over at an altitude of 51,140 ft, and
the average wind and actual wind profiles cross over at 48,980 ft. The total distance covered in
the actual wind case is found to be larger (by about 8%) than the distance covered in the constant
average and no wind cases. The maximum distance covered in the actual wind case is very close
to that of the average wind case, with the difference being about 1.4%. These qualitative
observations made in Figure 6 are also found to be valid for other values of 𝜒0 , as seen in
Figures 7 and 8. For reference, the no wind case for 𝜒0 = 0o has also been shown.
and 8, respectively. The total distance covered is observed to be greater for a lower value of 𝜒0 ,
with 𝜒0 = 0o yielding the greatest value and 𝜒0 = 90o the least value of the maximum range, as
It is observed that the wind modifies the trajectory of the aircraft. After starting in the
specified initial direction, the trajectory deflects so as to maximize the tailwind for maximum
Figure 9. Ground Track for Different Initial Heading Angles with No Wind (left) and Actual
Wind (right)
16
The time evolutions of true air speed, lift-to-drag ratio, angle of attack and flight path angle
are shown in Figures 10, 11, 12 and 13, respectively, for wind cases (i)-(iii). It can be noted from
these figures that the curves for cases (ii) and (iii) are very close, indicating that the use of a
constant value of wind velocity matching the average value of the actual wind profile, gives good
estimates for these optimal trajectories. Some interesting additional observations can be made
regarding Figures 10-13. There is a rapid decrease in the velocity of the aircraft initially as it
increases its energy altitude. During this phase, kinetic energy is traded off for potential energy,
enabling the aircraft to obtain a greater glide range. This is followed by a constant velocity
phase, during which the lift-to-drag ratio increases to a maximum, with a simultaneous smooth
decrease in the angle of attack. Then comes a smooth, almost linear decrease in velocity, during
which time the lift-to-drag ratio remains at its maximum, and the angle of attack remains
constant. This maximum lift-to-drag ratio provides the smallest possible glide angle for
maximum possible range. During the landing phase, the velocity again rapidly decreases to a
final value observed to be close to the calculated stall velocity of 235.3 ft/s, determined using
classical aerodynamic theory26. The value of the angle of attack at this point is close to 0.35 rad,
the given value of αmax.. The above is observed to be true for all cases of 𝜒0 . There is no
significant variation in the flight path angle time history with wind profile or 𝜒0 , as seen in
Figure 14. These figures also show that there are durations in which true air speed, lift-to-drag
ratio, angle of attack and flight path angle remain nearly constant.
VI. Conclusion
The powered-off flight (glide) trajectories of a supersonic McDonnell Douglas F-4 Phantom
II were optimized for maximum range, using the Chebyshev Pseudospectral Method, in the
17
presence of different wind profiles and for different values of initial heading angle. The presence
of wind affects the path travelled by the aircraft and the total distance covered. The trajectories as
well as the state and controls variable time histories for the actual wind and constant average
wind profiles were found to match in most cases or were sufficiently close to each other. Hence,
it can be concluded that the assumption of a wind profile with constant wind velocity matching
the average wind velocity of the actual variable wind profile, will lead to reasonably accurate
predictions of optimal trajectories, state and control variables for maximum range glide
It was also found that optimal glide trajectories of the McDonnell Douglas F-4 Phantom II
aircraft consist of four distinct phases, including (i) an initial climb phase, during which there is
an increase in altitude and a rapid decrease in velocity as trade-off between kinetic and potential
energy, as expected, followed by (ii) a constant velocity phase (consistent with predictions from
classical aerodynamics), (iii) a maximum lift-to-drag ratio phase, and (iv) a landing phase, during
Future work will include determination of optimal trajectories, state and control variables for
minimum time and minimum fuel for a supersonic aircraft in climb, in the presence of wind, with
realistic wind profiles. Optimum trajectories for various maneuvers will also be determined for
aircraft significantly different from the one considered in this paper, such as UAVs and MAVs.
18
References
1
Bliss G. A., Lectures on the Calculus of Variations, The Univ. of Chicago Press, 1946
2
Pontyragin L. S., The Mathematical Theory of Optimal Processes, New York, Intersciences,
1962
3
Bryson, A. E., and Ho, Y. C., Applied Optimal Control, Blaisdell Publishing Company,
Guidance, Control, and Dynamics, Vol. 21, No. 2, 1998, pp. 193-207
6
Betts, J. T., and Hallman, W. P., “NLP2 Optimization Algorithm Documentation,” The
1997
7
Rosen, J. B., and Kreuser, J. L., “A Gradient Projection Algorithm for Nonlinear
edited by H. J. Greenberg, Sijthoff and Noordhoff, Alphen aan den Rijn, The Netherlands, 1978,
pp. 335–362
9
Bryson, A. E., and Denham, W. F., “A Steepest-Ascent Method for Solving Optimum
Programming Problems,” Journal of Applied Mechanics, Vol. 29, 1962, pp. 247–257
10
Denham, W. F., Steepest Ascent Solutions of Optimal Programming Problems, Ph.D
19
11
Bryson Jr., A. E., and Desai, M. N., “Energy-State Approximation in Performance
Optimization of Supersonic Aircraft”, Journal of Aircraft, Vol. 6, No. 6, 1969, pp. 481-488
12
Fahroo, F., and Ross I. M., “Direct Trajectory Optimization by a Chebyshev Pseudospectral
Method”, Journal of Guidance, Control, and Dynamics, Vol 25, No. 1, 2002, pp. 160-166
13
Dai, R., and Cochran Jr., J. E., “Three-Dimensional Trajectory Optimization in Constrained
Event of Engine Cut-off”, Technion - Israel Institute of Technology Haifa, Israel, AIAA
Guidance, Navigation, and Control Conference, 08-11 August 2011, Portland, Oregon, AIAA-
2011-6596
15
Trefethen, L. N., Spectral Methods in MATLAB, Society for Industrial and Applied
MATLAB Driver for the SNOPT Optimization Software,” Proceedings of the 42nd
Germany, 1989
20
20
Hargraves, C. R., and Paris, S.W., “Direct Trajectory Optimization Using Nonlinear
Programming and Collocation,” Journal of Guidance, Control, and Dynamics, Vol. 10, No. 4,
Programming, Society for Industrial & Applied Mathematics, 2nd Edition, 2009
23
U.S. Standard Atmosphere, 1976, U.S. Government Printing Office, Washington, D.C.,
1976
24
Range Reference Atmosphere Committee, “Range Reference Atmosphere 0-70 km, Cape
Canaveral, Florida,” Range Commanders Council, Doc. 361-83, White Sands, NM, Feb. 1983
25
Adelfang, S. I., Smith, O. E., and Batts, G. W., “Ascent Wind Model for Launch Vehicle
Design”, Journal of Spacecraft and Rockets, Vol 31, No. 3, 1994, pp. 502-508
26
Yechout, T. R., Morris, S. L., Bossert, D. E., and Hallgreen, W. F., Introduction to Aircraft
21