Analysis of Hypersonic Flight Dynamics With Probabilistic Uncertainty in System Parameters

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

AIAA Guidance, Navigation and Control Conference and Exhibit AIAA 2008-6814

18 - 21 August 2008, Honolulu, Hawaii

Analysis of Hypersonic Flight Dynamics with


Probabilistic Uncertainty in System Parameters
∗ †
Avinash Prabhakar Raktim Bhattacharya
Aerospace Engineering, Texas A&M University
College Station, Texas, 77843-3141

In this paper we present a novel computational framework for analyzing evolution of


uncertainty in state trajectories of a hypersonic air vehicle due to uncertainty in initial
conditions and other system parameters. The framework is built on the so called gener-
alized polynomial chaos expansions. In this framework, stochastic dynamical systems are
transformed into equivalent deterministic dynamical systems in higher dimensional space.
In the research work presented here we study evolution of uncertainty due to initial condi-
tion, ballistic coefficient, lift over drag ratio and atmospheric density. The problem studied
here is related to the Mars entry descent landing problem.

I. Introduction
NASA has clearly identified the need for fundamental research on entry, descent, and landing of large robotic
and manned spacecraft (> 30 MT) on the surface of Mars with high accuracy. The expected mass of the
next Mars Science Laboratory mission is approximately 2, 800 kg at entry. The mission includes plans for a
precision guided entry and a tether-based payload deployment system, which is expected to provide a land-
ing accuracy of 20 − 40 km, from the target. Another major concern with high-mass entry is the mismatch
between the entry conditions and the deceleration capabilities provided by supersonic parachute technolo-
gies. In such applications, there is uncertainty present in initial condition and other system parameters.
Hence, for successful mission, it is critical to study the impact of such uncertainties on state trajectories
and determine uncertainty in the landing site and the entry condition for super sonic parachute deployment.
Figure 1 summarizes the research problem addressed in this paper.

The problem of studying the evolution of uncertainty in dynamical systems has been of interest consistently
in the scientific community. In many cases, the uncertainty in initial condition and other system parameters
for a dynamical system are not known precisely. However, characteristics of the uncertainty, such as mean,
standard deviation, probability density function, etc., may be known. A common technique to study the
evolution of such distributions is the Monte-Carlo simulation. In this approach, a set of initial conditions
that best represent the distribution, are propagated over the desired time interval, and statistical properties
of the envelope of trajectories are studied. For example, the expected value of the set of trajectories may
be evaluated easily, once all the trajectories are known. However, Monte-Carlo simulations are essentially
brute force in nature, with the understanding that a large number of points may be required to accurately
represent the distribution on initial conditions. As a consequence, the time required for computation is
greatly affected by the dimension of the system and the desired accuracy of the result. Other alternatives
include Gaussian closure of the parametric uncertainty and analysis of its impact by means of linear analysis.
For nonlinear systems, this is achieved by continuous linearization. Such methods are known not to work
∗ Graduate Student, [email protected]
† Assistant Professor, [email protected]

Copyright © 2008 by Avinash Prabhakar. Published by the American Institute of Aeronautics and Astronautics, Inc., with permission.
(a) Uncertainty in initial condition and its impact on state tra- (b) Uncertainty in the landing site.
jectories.
Figure 1. Uncertainty evolution in hypersonic flight dynamics (Image: courtesy NASA).

well when nonlinearities are significant.

In this research effort we are interested in developing methods for uncertainty propagation based on polyno-
mial chaos expansions, which can be thought of as an extension of Volterra’s theory of nonlinear functionals
for stochastic systems.1 Polynomial chaos (PC) was first introduced by Wiener2 where Hermite polynomi-
als were used to model stochastic processes with Gaussian random variables. According to Cameron and
Martin3 such an expansion converges in the L2 sense for any arbitrary stochastic process with finite second
moment. This applies to most physical systems. Xiu et al.4 generalized the result of Cameron-Martin to
various continuous and discrete distributions using orthogonal polynomials from the so called Askey-scheme5
and demonstrated L2 convergence in the corresponding Hilbert functional space. This is popularly known
as the generalized polynomial chaos (gPC) framework. The gPC framework has been applied to various
applications including stochastic fluid dynamics,6, 7 stochastic finite elements,8 and solid mechanics.9, 10 Ap-
plication of gPC to problems related to control and estimation of dynamical systems, has been surprisingly
limited.

The paper is organized as follows. We first present preliminaries on the theory of polynomial chaos and
demonstrate transformation of stochastic dynamics, with parametric uncertainty, into deterministic dynamics
in higher dimensional state space. We then present the stochastic Vinh’s equation11 for longitudinal motion
where we assume uncertainty in ballistic coefficient, lift over drag ratio, density parameters and initial
conditions. The stochastic hypersonic dynamics is then transformed into deterministic dynamics in higher
dimensional state space using polynomial chaos expansions. This is followed by numerical results that
characterize uncertainty propagation in state trajectories induced by uncertainty in system parameters. The
results obtained from polynomial chaos framework are compared with Monte-Carlo simulations, which are
shown to agree well.

II. Polynomial Chaos


II.A. Generalized Polynomial Chaos
Let (Ω, F, P ) be a probability space, where Ω is the sample space, F is the σ-algebra of the subsets of Ω,
and P is the probability measure. Let ∆(ω) = (∆1 (ω), · · · , ∆d (ω)) : (Ω, F) → (Rd , B d ) be an Rd -valued
continuous random variable, where d ∈ N, and B d is the σ-algebra of Borel subsets of Rd . A general second

2
order process X(ω) ∈ L2 (Ω, F, P ) can be expressed by polynomial chaos as

X
X(ω) = xi φi (∆(ω)), (1)
i=0

where ω is the random event and φi (∆(ω)) denotes the gPC basis of degree p in terms of the random variables
∆(ω). The functions {φi } are a family of orthogonal basis in L2 (Ω, F, P ) satisfying the relation

E[φi φj ] = E[φ2i ]δij , (2)

where δij is the Kronecker delta and E[·] denotes the expectation with respect to the probability measure
dP (ω) = f (∆(ω))dω and probability density function f (∆(ω)). Henceforth, we will use ∆ to represent ∆(ω).

For random variables ∆ with certain distributions, the family of orthogonal basis functions {φi } can be
chosen in such a way that its weight function has the same form as the probability density function f (∆).
These orthogonal polynomials are members of the Askey-scheme of polynomials,5 which form a complete
basis in the Hilbert space determined by their corresponding support. Table 1 summarizes the correspon-
dence between the choice of polynomials for a given distribution of ∆.4

Random Variable ∆ φi (∆) of the Wiener-Askey Scheme


Gaussian Hermite
Uniform Legendre
Gamma Laguerre
Beta Jacobi
Table 1. Correspondence between choice of polynomials and given distribution of ∆(ω).

II.B. Approximate Solution of Stochastic Differential Equations


A stochastic dynamical system of the form ẋ = f (x, ∆), where x ∈ Rn , ∆ ∈ Rd , can be solved using the
polynomial chaos framework in the following manner. Assume solution of the stochastic differential equation
to be x(t, ∆). For second order processes, the solution for every component of x ∈ Rn can be approximated
as
P
X
x̂i (t, ∆) = xij (t)φj (∆); i = 1, · · · , n. (3)
j=0

Substituting the approximate solution into the dynamical system results in error we get,

ei = x̂˙ i − fi (x̂, ∆); i = 1, · · · , n.

The approximation in eqn.(3) is optimal in the L2 sense when the projections of the error on the orthogonal
basis functions are zero, i.e.,
Z
hei (t, ∆), φj (∆)i = ei (t, ∆)φj (∆)f (∆)d∆ = 0; j = 0, · · · , P ; i = 1, · · · , n; (4)
D∆

where D∆ is the domain of ∆. Equation, eqn.(4) results in n(P + 1) deterministic ordinary differential
equations, which can be solved numerically to obtain the approximated stochastic response. Therefore,
the stochastic dynamics in Rn has been transformed into deterministic dynamics in Rn(P +1) . The series is
truncated after P + 1 terms, which is determined by the dimension d of ∆ and the order r of the orthogonal
polynomials {φj }, satisfying P + 1 = (d+r)!
d!r! .

3
III. Vinh’s Equation with Probabilistic System Parameters
In this section we use polynomial chaos framework for analyzing uncertainty propagation in hypersonic flight
vehicles. Here we consider 3DOF Vinh’s equation to model hypersonic flight dynamics, the state variables
are height h from surface of planet, velocity v and flight path angle γ. The equation of motion in the
longitudinal plane is given by

ḣ = v sin γ, 


ρ0 h2 −hR0
v̇ = − 2Bcexp(  h1 )v 2 − sin γ, (5)
2
γ̇ = cos γ v −1 + ρ0 ν exp( h2 −hR0 )v, 


v 2Bc h1

where R0 , h1 , h2 , Bc , ρ0 and ν are constants given by the following table.

Description Symbol & Value


Density on the surface of Mars ρ0 = 0.0176kg/m3
Gravitational Constant of Mars: µ = 42.828 × 103 Km/s2
Scale Height in Density Computation h1 = 9.8 × 103 m, h2 = 20 × 103 m
Mean Equatorial radius of Mars R0 = 3397Km
Ballistic Coefficient Bc = 72.8kg/m2
Gravitational Oblateness Coeff. for Mars J2 = 1960.45 × 10−6
Lift to Drag Ratio ν = 0.3
Surface Gravity on Mars gm = 3.71
Table 2. Parameters for a Mars atmospheric entry vehicle.

We consider uncertainty in parameters ρ0 (density on surface of Mars), ν (lift over drag ratio) and Bc (bal-
listic coefficient). The uncertainty in these parameters are assumed to be uniform and we are interested in
determining the uncertainty in state trajectories in flight and at the time when the vehicle strikes the ground.

When ν, Bc and ρ0 are random variables, the differential equation defined by eqn.(5) is stochastic, and h,
v, and γ are random processes. We use polynomial chaos framework to transform the stochastic differential
equation into a deterministic differential equation in higher dimension state space. It is assumed that
random variables ν, Bc and ρ0 are expressed in terms of probability density functions. In this paper we
assume uniform distribution on ν, Bc and ρ0 . Therefore, from table(1), the basis functions are given by
Legendre polynomials. The random variables ν, Bc and ρ0 can then be written as,

ν(∆) = ν̄ + δν∆,
Bc (∆) = B̄c + δBc ∆,
ρ0 (∆) = ρ̄0 + δρ0 ∆,

where ∆ ∈ [−1, 1] and δν, δBc , and δρ0 are the perturbations about nominal values ν, B̄c and ρ̄0 respectively.
In this research we have assumed the parameters to have uniform distributions about their nominal values.
The random processes h(t, ∆), v(t, ∆) and γ(t, ∆) are expanded as
PP
h(t, ∆) = h (t)φi (∆),
P0P i
v(t, ∆) = v (t)φi (∆),
P0P i
γ(t, ∆) = 0 γi (t)φi (∆).

Note that for the parameters with uniform uncertainty, only two basis functions are required to completely
capture their respective probability density functions. No benefit is obtained by including more terms.

4
However for the states, the expansion includes higher order basis functions. From the gPC theory, we are
guaranteed exponential convergence as higher order basis functions are included.4

Substituting these in eqn.(5) results in the following


PP PP PP
0 ḣi φi = 0 vi φi sin( 0 γi φi ),
!
h
PP ρ0 (∆) exp h2 PP PP PP
0 v̇i φi = − 2Bc (∆) R0
PP
1
hi φi 0 0 φi φj vi vj − sin( 0 γi φi ),
exp 0
h1

h2
PP PP PP ρ0 (∆)ν(∆) exp PP
PP 1
h1
0 γ̇i φi = cos( 0 γi φi ) ( 0 vi φ i − ) + 2Bc (∆) (
PP ) 0 vi φ i .
0 vi φi
R0 0 hi φi
exp h1

The above equations are further simplified by the following substitutions


PP
0 wi φi = PP 1v φ ,
0 i i

h2
PP exp h1
0 x i φi = R0
PP ,
exp 0 hi φi
h1

PP ρ0 (∆)
0 yi φi = 2Bc (∆) ,
PP ρ0 (∆)ν(∆)
0 zi φi = 2Bc (∆) ,

where coefficients wi , xi , yi , zi ∈ R are yet to be determined. The Vinh’s equation can now be written as
PP PP P 
P
0 ḣi φi = 0 vi φi sin 0 γi φi ,

PP PP P 
P
0 v̇i φi = − i,j,k,l=0 φi φj φk φl xi yj vk vl − sin i=0 γi φi ,

PP P  P  P
P P PP P
0 γ̇ i φ i = cos i=0 γ i φ i i=0 v i φ i − i=0 w i φ i + i,j,k=0 φi φj φk xi zj vk .

Taking Galerkin projection on basis functions φi (∆), we get the following deterministic differential equations,
PP D P
P
E 
ḣm = hφ12 i 0 vi φi φm sin 0 γi φi , 

m 

D  E 

P P

1 1
P P
v̇m = − hφ2 i i,j,k,l=0 hφi φj φk φl φm i xi yj vk vl − hφ2 i φm sin γ φ ,

i i

m m i=0 
D P E D P E  (6)
PP P PP P
γ̇m = hφ12 i i=0 vi φi φm cos γ
i=0 i i φ − 1
hφ2m i w
i=0 i φ φ
i m cos γ φ
i=0 i i



m 



1
P P 
+ hφ2 i i,j,k=0 hφi φj φk φm i xi zj vk , 
m

where m = 0, · · · , P . Equation (6) is the equivalent deterministic dynamics of the stochastic dynamics
given by eqn.(5), approximated by gPC expansions. Solution of this differential equation, in higher dimen-
sional state space, can then be used to characterize h(t, ∆), v(t, ∆) and γ(t, ∆).

The terms wi , xi , yi and zi in eqn.(6) are computed as follows. Define


P PP
X R0 0 hi φi
α= γi φi , β = .
0
h1

Multiplying,
P
X exp hh12
x i φi = PP ,
R0 hi φi
0 exp h1
0

5
by exp(β) on both the sides we get
P
h2 X
exp = xi φi exp β.
h1 0
Taking projection on the basis functions yields,
P
h2 X
hexp φk i = h φi φk exp βixi .
h1 0

This produces a set of (P + 1) linear equations in xi . Similarly, wi are computed by multiplying


P
1 X
PP = w i φi ,
0 vi φ i 0
PP
by 0 vi φi on both the sides, which reduces to
P X
X P
1= φi φj vi wj .
0 0

Now taking projection on the basis functions we get,


P X
X P
hφk i = hφi φj φk ivi wj ,
0 0

which also is a system of linear equations in wi . The coefficients yi and zi can also be determined in the
similar manner.

The terms sin(α) and exp(β) are computed numerically. These terms can also be computed by approximating
them by using Taylor series expansion of the perturbation about the mean.12 For example, computing the
exponential of a random variable ξ, the perturbation around the mean is given as d = ξ − ξ0 , where ξ0 is the
mean. Therefore,
Ntay n
X d
exp(ξ) = exp(ξ0 )(1 + ).
n=1
n!
Here d is the stochastic part of ξ and Ntay is the number of terms in the Taylor series expansion. The sine
of the random process is computed similar to the exponential, in the following manner,
sin(ξ) = sin(ξ0 + d) = sin(ξ0 ) cos(d) + cos(ξ0 ) sin(d).
While Taylor series approximation is straightforward and generally computationally cost effective, it can be-
come severely inaccurate when higher order gPC expansions are required to represent the physical variability.

To tackle the problem of inaccuracies in the evaluation of transcendental functions, using Taylor series
expansions, a more robust algorithm is presented by Debusschere et al.12 This method is valid for any non
polynomial function u(ξ) for which du
dξ can be expressed as a rational function of ξ, u(ξ). This approach has
not been applied in the research presented in this paper.

IV. Obtaining Statistics from Polynomial Chaos


Mean and variance of the state trajectories can be easily computed from the coefficients of the gPC expan-
sions. The mean trajectories can be derived as,
Xp p
X p
X Z
E[h(t, ∆)] = E[ hi φi ] = hi E[φi ] = hi φi f d∆,
i=0 i=0 i=0 D∆

6
and similarly,
Pp R
E[v(t, ∆)] = i=0 vi D∆
φi f d∆,
Pp R
E[γ(t, ∆)] = i=0 γi D∆
φi f d∆,
where hi , vi , γi are the coefficients of the gPC expansions of h(t, ∆), v(t, ∆), γ(t, ∆) respectively, and f is the
probability density function of the parameters.

The variance of a trajectory h(t, ∆) is given by

σ 2 [h(t, ∆)] = E[(h(t, ∆)2 ] − E[h(t, ∆)]2 ,


PP R 2
= i,j=0 hi hj D∆ φj φj f d∆ − E[h(t, ∆)]
PP 2
R 2 2
= i=0 hi D∆ φi f d∆ − E[h(t, ∆)] , because of orthogonality of φi , φj .

In this manner, the covariance matrix for the system can also be determined.

V. Simulation Results
In this paper, we analyze the effect of uncertainty in initial condition and parameters on hypersonic flight
trajectories and compare the results obtained from Monte-Carlo simulations and polynomial chaos theory.
We have assumed 10% parametric variation in initial conditions (h0 , v0 , γ0 ) and parameters (ρ0 , ν, Bc), about
the nominal values. The nominal values for (h0 , v0 , γ0 ) are taken as (100 km, 3.5 km/s, −2 deg) respectively.
The nominal values for (ρ0 , ν, Bc) are listed in listed in table(2).The distribution is assumed to be uniform
about the nominal values. To identify the sensitivity of the state trajectories to these uncertain parameters,
we have performed the analysis with uncertainty present in one parameter at a time. A coupled analysis
can also be performed, by considering all the parameters to be uncertain at the same time. We have ob-
tained statistics of the states trajectories from Monte-Carlo simulations and have compared them with those
obtained from polynomial chaos theory. The results match very well and are obtained with significantly
reduced computational overhead.

Figure (2) shows the evolution of uncertainty in state trajectories due to parametric uncertainty. We observe
that the effect of uncertainty in ρ0 , ν and Bc are pronounced in lower altitudes where the aerodynamic effects
are dominant. Figure (3) shows the results for uncertainty in initial conditions. We observe that the state
trajectories of the vehicle are greatly affected by the uncertainty in initial condition. These trajectories were
obtained using Monte-Carlo simulations.

We next analyze the sensitivity of the state trajectories with respect to the above mentioned parametric
uncertainty. From figures 4,5,6,7,8,9, we observe that the mean trajectories are almost identical for the various
parametric uncertainty considered. There is however significant difference in the variance trajectories. Figure
(10(a)) shows the convergence of error in the statistics obtained from Monte-Carlo method and Polynomial
Chaos theory. We see that better results are obtained for higher order approximations. Figure (10(b)) shows
the rise in computational time for higher order polynomial chaos expansions, this is however significantly
less when compared to the time taken by Monte-Carlo methods. For the result presented here, Monte-Carlo
simulations took 116.12 seconds.

VI. Summary and Conclusions


In this paper, we have presented a framework for analyzing evolution of uncertainty in hypersonic flight.
This framework is based on the generalized polynomial chaos framework, which has been successfully applied
to areas such as stochastic computational fluid dynamics and solid mechanics. We have demonstrated that

7
this framework captures the trajectory statistics quite accurately and is computationally less demanding
than methods based on Monte-Carlo simulations. Monte-Carlo methods were used in this paper solely to
verify the results obtained from the generalized polynomial chaos approach.

However, the generalized polynomial chaos based approach has limitations related to errors due to long
term integration. This makes it suitable for estimation of short term statistics. Methods exist that reduce
the error growth over time and they incorporate adaptive multi-element schemes.13, 14 The multi-element
approach approximates Ω using local approximations, instead of global approximations. It is well known in
approximation theory that locally supported basis functions provide superior results than globally supported
functions. We will incorporate these approaches in our future research.

VII. Acknowledgements
This research work was conducted under NASA funding from NRA NNH06ZEA001N-HYP, under cooperative
agreement number NNX07AC44A: Fundamental Aeronautics: Hypersonics Project, Topic 5.4 Advanced
Control Methods, Subtopic A.5.4.1 Advanced Adaptive Control, with Dan Moerder from NASA Langley as
the technical monitor.

References
1 V. Volterra. Lecons sur les Equations Integrales et Integrodifferentielles. Paris: Gauthier Villars, 1913.
2 N. Wiener. The Homogeneous Chaos. American Journal of Mathematics, 60(4):897–936, 1938.
3 R. H. Cameron and W. T. Martin. The Orthogonal Development of Non-Linear Functionals in Series of Fourier-Hermite

Functionals. The Annals of Mathematics, 48(2):385–392, 1947.


4 Dongbin Xiu and George Em Karniadakis. The Wiener–Askey Polynomial Chaos for Stochastic Differential Equations. SIAM

J. Sci. Comput., 24(2):619–644, 2002.


5 R. Askey and J. Wilson. Some Basic Hypergeometric Polynomials that Generalize Jacobi Polynomials. Memoirs Amer.

Math. Soc., 319, 1985.


6 Thomas Y. Hou, Wuan Luo, Boris Rozovskii, and Hao-Min Zhou. Wiener Chaos Expansions and Numerical Solutions of

Randomly Forced Equations of Fluid Mechanics. J. Comput. Phys., 216(2):687–706, 2006.


7 Dongbin Xiu and George Em Karniadakis. Modeling Uncertainty in Flow Simulations via Generalized Polynomial Chaos. J.

Comput. Phys., 187(1):137–167, 2003.


8 Roger G. Ghanem and Pol D. Spanos. Stochastic Finite Elements: A Spectral Approach. Springer-Verlag New York, Inc.,

New York, NY, USA, 1991.


9 Roger Ghanem and John Red-Horse. Propagation of Probabilistic Uncertainty in Complex Physical Systems Using a Stochas-

tic Finite Element Approach. Phys. D, 133(1-4):137–144, 1999.


10 R.G. Ghanem. Ingredients for a General Purpose Stochastic Finite Elements Implementation. Comput. Methods Appl. Mech.

Eng., 168(1-4):19–34, 1999.


11 Nguyen Vinh. Hypersonic and Planetary Entry Flight Mechanics. Univ of Michigan Press, June 1980.
12 Bert J. Debusschere, Habib N. Najm, Philippe P. Pébay, Omar M. Knio, Roger G. Ghanem, and Olivier P. Le Maı̂tre.

Numerical Challenges in the Use of Polynomial Chaos Representations for Stochastic Processes. SIAM J. Sci. Comput.,
26(2):698–719, 2005.
13 R. Li and R. Ghanem. Adaptive Polynomial Chaos Expansions Applied to Statistics of Extremes in Nonlinear Random

Vibration. Probabilistic Engineering Mechanics, 13:125–136, April 1998.


14 Xiaoliang Wan and George Em Karniadakis. An Adaptive Multi-Element Generalized Polynomial Chaos Method for Stochas-

tic Differential Equations. J. Comput. Phys., 209(2):617–642, 2005.

8
(a) Height vs Time (b) Velocity vs Time (c) Flight path angle vs Time

Figure 2. Evolution of uncertainty with 5% uniform uncertainty in ν,Bc and ρ0 . Blue shows the various trajectories
obtained for uncertain ρ0 , ν and Bc . The black line shows the mean of these trajectories.

0.035 1.4 0.2

0.03 0
1.2

0.025
−0.2
1
0.02
−0.4
0.015 0.8

Gamma
H

−0.6
0.01 0.6
−0.8
0.005
0.4
−1
0

0.2 −1.2
−0.005

−0.01 0 −1.4
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
T T T

(a) Height vs Time (b) Velocity vs Time (c) Flight Path Angle vs Time

Figure 3. Evolution of uncertainty with 5% uniform uncertainty in initial conditions. Blue lines show the various
trajectories obtained for uncertain h(t0 ), v(t0 ) and γ(t0 ). The black line shows the mean of these trajectories.

9
4
x 10
10 4000 0 12 25 0.06
8 10 0.05

Variance velocity (m/s)

Variance gamma (rad)


−0.2 20

Mean velocity (m/s)

Mean gamma (rad)

Variance height (m)


3000
Mean height (m)

6 8 0.04
−0.4 15
4 2000 6 0.03
−0.6 10
2 4 0.02
1000
0 −0.8 5
2 0.01

−2 0 −1 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t
−4 −4
x 10 x 10
20 0.25 8 0.025 0.06

Error in variance velocity (m/s)

Error in variance gamma (rad)


Error in mean velocity (m/s)

Error in mean gamma (rad)

Error in variance height (m)


Error in mean height (m)

0.2 0.02 0.05


15 6
0.04
0.15 0.015
10 4 0.03 1
0.1 0.01
0.02
5 2
0.05 0.005 0.01

0 0 0 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t

(a) Mean Trajectories (b) Variance Trajectories

Figure 4. 10% uncertainty in h0 , Blue (dashed) = Monte-Carlo, Red (solid) = Polynomial Chaos.

4
x 10
12 4000 0
2000 800 0.4
10 3500
Mean velocity (m/s)

Mean gamma (rad)

−0.2

Variance velocity (m/s)


Mean height (m)

Variance gamma (rad)


Variance height (m)

3000 1500 600 0.3


8
2500 −0.4
6 1000 400 0.2
2000
−0.6
4 1500 500 200 0.1

2 1000 −0.8
0 500 1000 0 500 1000 0 500 1000 0 0 0
t t t 0 500 1000 0 500 1000 0 500 1000
t t t
−3
x 10 −3
40 1.5 3 x 10
3 2.5 4
Error in mean velocity (m/s)

Error in mean gamma (rad)

Error in variance velocity (m/s)

Error in variance gamma (rad)


Error in mean height (m)

Error in variance height (m)

2.5
30 2.5 2
1 2 3
2
1.5
20 1.5
1.5 2
0.5 1 1
1
10 1
0.5 0.5 0.5

0 0 0 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t

(a) Mean Trajectories (b) Variance Trajectories

Figure 5. 10% uncertainty in v0 , Blue (dashed) = Monte-Carlo, Red (solid) = Polynomial Chaos.

10
4
x 10
15 4000 0
2.5 6 0.015

−0.2 5

Variance velocity (m/s)

Variance gamma (rad)


Mean velocity (m/s)

Mean gamma (rad)


10 3000 2

Variance height (m)


Mean height (m)

−0.4 4 0.01
1.5
5 2000 3
−0.6 1
2 0.005
0 1000
−0.8 0.5 1
−5 0 −1 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t
−4 −3 −5
x 10 x 10 x 10
15 0.2 8 4 0.01 8

Error in variance velocity (m/s)

Error in variance gamma (rad)


Error in mean velocity (m/s)

Error in mean gamma (rad)

Error in variance height (m)


Error in mean height (m)

0.008
0.15 6 3 6
10
0.006
0.1 4 2 4
0.004
5
0.05 2 1 2
0.002

0 0 0 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t

(a) Mean Trajectories (b) Variance Trajectories

Figure 6. 10% uncertainty in γ0 , Blue (dashed) = Monte-Carlo, Red (solid) = Polynomial Chaos.

4
x 10
15 4000 0 x 10
−4
0.05 0.4 8
−0.2
Mean velocity (m/s)

Mean gamma (rad)

Variance velocity (m/s)

10 3000
Variance gamma (rad)

0.04
Mean height (m)

Variance height (m)

0.3 6
−0.4
5 2000 0.03
−0.6 0.2 4
0.02
0 1000
−0.8 0.1 2
0.01

−5 0 −1
0 500 1000 0 500 1000 0 500 1000 0 0 0
0 500 1000 0 500 1000 0 500 1000
t t t
t t t
−3
x 10 −4 −3 −5
20 0.2 1 x 10 x 10 x 10
8 1.2 3
Error in variance velocity (m/s)

Error in variance gamma (rad)


Error in mean velocity (m/s)

Error in mean gamma (rad)

Error in variance height (m)


Error in mean height (m)

0.8 1 2.5
15 0.15 6
0.6 0.8 2
10 0.1 4 0.6 1.5
0.4
0.4 1
5 0.05 2
0.2
0.2 0.5

0 0 0 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t

(a) Mean Trajectories (b) Variance Trajectories

Figure 7. 10% uncertainty in Bc , Blue (dashed) = Monte-Carlo, Red (solid) = Polynomial Chaos.

11
4
x 10
15 4000 0 5 6 0.025

Variance velocity (m/s)


−0.2

Variance gamma (rad)


Mean velocity (m/s)
4 0.02

Mean gamma (rad)

Variance height (m)


10 3000
Mean height (m)

4
−0.4 3 0.015
5 2000 3
−0.6 2 0.01
2
0 1000
−0.8 1 0.005
1

−5 0 −1 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t
−4 −4
x 10 x 10
15 0.2 6 0.012 0.012 1

Error in variance velocity (m/s)

Error in variance gamma (rad)


Error in mean velocity (m/s)

Error in mean gamma (rad)

Error in variance height (m)


Error in mean height (m)

5 0.01 0.01 0.8


0.15
10 4 0.008 0.008
0.6
0.1 3 0.006 0.006
0.4
5 2 0.004 0.004
0.05
1 0.002 0.002 0.2

0 0 0 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t

(a) Mean Trajectories (b) Variance Trajectories

Figure 8. 10% uncertainty in ν, Blue (dashed) = Monte-Carlo, Red (solid) = Polynomial Chaos.

4
x 10
15 4000 0 −4
x 10
0.05 0.4 8
−0.2
Mean velocity (m/s)

Mean gamma (rad)

10 3000
Variance velocity (m/s)

Variance gamma (rad)


Mean height (m)

0.04
Variance height (m)

0.3 6
−0.4
5 2000 0.03
−0.6 0.2 4
0.02
0 1000
−0.8 0.1 2
0.01
−5 0 −1
0 500 1000 0 500 1000 0 500 1000 0 0 0
t t t 0 500 1000 0 500 1000 0 500 1000
t t t
−3
x 10 −4 −3 −5
20 0.2 1 x 10 x 10 x 10
8 1.2 3
Error in mean velocity (m/s)

Error in variance velocity (m/s)


Error in mean gamma (rad)

Error in variance gamma (rad)


Error in mean height (m)

Error in variance height (m)

0.8 1 2.5
15 0.15
6
0.6 0.8 2
10 0.1
4 0.6 1.5
0.4
0.4 1
5 0.05 2
0.2
0.2 0.5

0 0 0 0 0 0
0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000 0 500 1000
t t t t t t

(a) Mean Trajectories (b) Variance Trajectories

Figure 9. 10% uncertainty in ρ0 , Blue (dashed) = Monte-Carlo, Red (solid) = Polynomial Chaos.

12
−4
x 10
12.7 0.1758 6.33
Max error in mean velocity (m/s)
Max error in mean height (m)

4.5
Max error in mean gamma

0.1756 6.325
12.68
0.1754
6.32
12.66 0.1752
4
6.315
0.175
12.64
0.1748 6.31
Computational time (sec)

12.62 0.1746 6.305 3.5


0 10 20 0 10 20 0 10 20
Order of gPC expansion Order of gPC expansion Order of gPC expansion

−3 −3 −5
x 10 x 10 x 10
3.7 9.6 6.05
3
Max error in var velocity

Max error in var gamma

3.6 9.4 6
Max error in var height

3.5 9.2 5.95

3.4 9 5.9
2.5
3.3 8.8 5.85

3.2 8.6 5.8

3.1 5.75
0 10 20 0 10 20 0 10 20 2
2 4 6 8 10 12 14 16
Order of gPC expansion Order of gPC expansion Order of gPC expansion
Order of gPC expansion

(a) Error Convergence (b) Computational Time


Figure 10. Error convergence and computational time

13

You might also like