International Journal of Thermal Sciences 118 (2017) 497e509

Contents lists available at ScienceDirect

International Journal of Thermal Sciences

journal homepage: www.elsevier.com/locate/ijts

Thermochemical ablation modeling forward uncertainty

analysisdPart I: Numerical methods and effect of model parameters
Alessandro Turchi a, *, Pietro M. Congedo b, Thierry E. Magin a
a
von Karman Institute for Fluid Dynamics, B-1640 Rhode-St-Gen

ese, Belgium
b
Inria Bordeaux e Sud-Ouest, Team CARDAMOM, 33405 Talence, France

a r t i c l e i n f o a b s t r a c t

Article history: Next generation spacecraft will bring back heavier payloads from explored planets. Advance in the
Received 26 December 2015 modeling of the thermo-chemical ablation of carbon-based thermal protection system materials is
Received in revised form fundamental to improve the design capabilities of these vehicles. Computational fluid dynamic ap-
21 November 2016
proaches are extensively used to model the gas-surface interaction phenomena over ablative materials.
Accepted 3 April 2017
Available online 26 April 2017
The advantage of such kind of approaches is the accurate description of the aerothermal environment
obtained through the full resolution of the mechanical, thermal, and chemical boundary layers that
develop over an ablative surface when exposed to a high-enthalpy flow. This paper is devoted to the
Keywords:
Thermal protection system materials
assessment of the uncertainties of such kind of thermo-chemical ablation model and to study their effect
Carbon ablators on the model final outcomes. A sphere of non-pyrolyzing carbon-based material, exposed to conditions
Gas-surface interaction similar to those of a typical plasma wind tunnel test, is the selected test case for the analysis. Two for-
Ablation modeling ward non-intrusive uncertainty quantification techniques are used to analyze the influence of the
Uncertainty analysis defined set of uncertain parameters on the estimate of steady-state mass blowing flux and surface
temperature. Our results show that for the selected conditions, and uncertainty ranges, the surface
nitridation reaction probability has the strongest impact on the model outcomes.
© 2017 Elsevier Masson SAS. All rights reserved.

1. Introduction
During the past decades space agencies have sent landers to
Mars, Venus, and Titan and brought back samples from the Moon,
the Sun, comets and asteroids. The common denominator of such
kind of probes is that they need protection from the severe aero-
thermodynamic heating generated flying across the planet's at-
mosphere, when the high kinetic energy of the spacecraft is
transformed into thermal energy [1]. Passive thermal protection
systems (TPSs) are commonly used to fulfill the critical task of
protecting the spacecraft during this last phase of the mission [2]. In
practice, materials capable to survive the harsh thermo-chemical
environment, and sustain the predicted heat load, are super-
imposed to the vital structure of the spacecraft to build-up a
physical barrier against the high-enthalpy impinging flow. This
barrier is commonly referred to as heat shield. For spacecraft that
have to bear severe entry conditions (e.g., velocity and peak heat
flux above 10 km/s and 10 W/cm2, respectively), engineers cannot
* Corresponding author.
E-mail address: turchi@vki.ac.be (A. Turchi).
prescind, since the early ages of space exploration, from using
ablative materials to build efficient heat shields [2, 3]. In this case,
the TPS material is sacrificed during the atmospheric entry, un-
dergoing a series of thermo-chemical and mechanical processes
that help to dissipate the incoming energy at the price of its
structural integrity [4]. For present and future space exploration
missions, advanced ablative TPSs can be mission enabling, signifi-
cantly impacting the mass of both scientific and instrumental
payloads. Therefore, it is of fundamental importance to advance
capabilities for their modeling, design and analysis.
Carbon-based materials have been the subject of numerous
studies, as they represent the most performing subclass of ablative
TPS materials. Numerous experimental efforts have been made to
understand and characterize the physical phenomena occurring
when these kind of materials are exposed to high-enthalpy flows
[5e11]. Simultaneously, a multiplicity of modeling approaches have
been developed to study the gas-surface interaction and the ma-
terial response [12]. Theoretical and numerical models have been
developed ad hoc to analyze specific phenomena [13e16], to study
particular conditions [17, 18], or to perform global analyses of the
TPS material behavior [19e23]. Besides, tools that make use of
computational fluid dynamic (CFD) techniques have set. In the CFD

http://dx.doi.org/10.1016/j.ijthermalsci.2017.04.004
1290-0729/© 2017 Elsevier Masson SAS. All rights reserved.
498 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
Nomenclature
q_ radout solid radiative heat flux, W=m2
q_ cond conductive heat flux in the solid, W=m2
q_ radnet net radiative heat flux, W=m2
k surface reaction velocity, m=s
h static enthalpy, J=kg
pC;eq: saturated carbon vapor pressure, Pa
R universal gas constant, 8.31 kJ=ðmol KÞ
m_ surface mass blowing flux, kg=ðm2 sÞ
Nr number of surface reactions
N number of species or dimension of multi-dimensional
function
T temperature, K
ydi diffusive velocity of species i, m=s
y velocity, m=s
Greek Symbols
u_ riw reaction-specific surface source term of
species i, kg=ðm2 sÞ
approach the ablation is treated as a purely surface phenomenon,
and its coupling with the external flow field is handled through
dedicated boundary conditions in the CFD simulation of the flow
field [24]. This tools have been used to analyze the TPS interaction
with the surrounding environment. Both charring and non-
charring materials, using either equilibrium or finite-rate surface
chemistry, have been simulated in a wide range of TPS material
applications (e.g., spacecraft heat shields, solid rocket nozzle ther-
mal protections, ground testing of thermal protection materials)
[25e33]. The ability of such models to inherently handle the
coupling between the operational aerothermodynamic environ-
ment and the TPS surface is certainly an asset when the final goal is
the numerical reproduction of a real mission, or experiment, to
assist the heat shield design. In addition, the possibility of coupling
a steady-state CFD model with a transient material response allows
to consider both the gas-surface interaction and the in-depth ma-
terial response [34, 35], with clear benefits in terms of accuracy,
particularly when dealing with the analysis of strongly unsteady
phenomena.
Numerical simulations are a powerful tool in modern engi-
neering. They allow to predict qualitatively and quantitatively the
behavior of generic systems. However, to ensure the reliability of
these predictions, a systematic and comprehensive treatment of
both the calibration and the validation processes of the developed
models is fundamental.
This should also include the quantification of the inherent
model uncertainties arising from: the physical simplifications made
to obtain a mathematical model representative of the complex
phenomena under investigation; the numerical approximations
due to the finite discretization used in the numerical solver to
approximate the solution of the mathematical model; the lack of
knowledge on some of the model parameters. In this context, the
growing field of uncertainty quantification (UQ) aims at developing
rigorous frameworks and reliable methods to characterize the
impact of these uncertainties on the prediction of the quantities of
interest (QoI). Investigations that make use of UQ techniques have
been embraced with growing enthusiasm in the recent years
[36, 37]. Practical applications of these techniques to assess the
capabilities of deterministic simulation tools range from simple
u_ iw overall surface source term of species i,kg=ðm2 sÞ
g reaction probability
l thermal conductivity, W=m K
f generic variable
r density kg=m3
s Stefan-Boltzmann constant, 5:67  108 W=ðm2 K4 Þ
ε integral emissivity
e% percent coefficient of variation
Subscripts
b back surface of the material
s solid phase
w gas-surface interface
i ith species or generic index
j generic index
Superscripts
r r th surface reaction
heat transfer problems [38] to hypersonic flight predictions
[39, 40], and have been shown to bring useful information to sci-
entists working in either model development or system design.
This study deals with the analysis of the coupling between a CFD
deterministic approach to the ablation modeling, and modern un-
certainty quantification techniques. In particular, the work focuses
on the numerical study of the forward-stagnation-point ablation of
a spherical sample of non-pyrolyzing carbon-based TPS material
exposed to a subsonic high-enthalpy flow. Our interest lies in the
uncertainties associated to the limited knowledge, or the intrinsic
variability, of the input quantities needed to perform the CFD
analysis. The deterministic analysis requires, in fact, the precise
specification of some model parameters for which typically only
limited information is available from experimental observations.
The goal is to analyze the model dependencies on some critical
parameters, and to quantify how these uncertainties affect the
quantitative results of the model.
The article is arranged as follows. Section 2 describes theoreti-
cally the ablative boundary conditions, analyzes the model un-
certainties and presents the deterministic CFD tool used for the
study. Section 3 introduces the basic theory of UQ analysis, dis-
cussing the two mathematical approaches selected for the study. In
section 4, the UQ analysis on the ablation modeling, applied to a
nominal CFD test case, is presented. Two different scenarios, with
modified sets of model input uncertainties, are analyzed. Moreover,
an investigation on the important aspect of the definition of the
input uncertainty distributions is carried out. Finally, section 5
summarizes the outcomes of the analyses and discusses the
possible perspectives of the presented approach.
2. Surface ablation modeling
The study of the gas-surface interaction by means of a CFD
approach requires the implementation of dedicated boundary
conditions. For the present analysis we focus on the case of a non-
porous carbon-based non-charring material (e.g., graphite, carbon/
carbon, etc…). Moreover the following assumptions are considered
[25, 41]: i) the surface ablation is a pure thermo-chemical process
(i.e., no material can be removed in condensed phase); ii) the solid
 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
conduction is considered at steady-state; iii) the surface reactions
are assumed to be uncoupled and first order in reactant concen-
trations. The derivation of the ablative boundary condition can be
found in several works [24, 42]. For the sake of clearness, we pre-
sent here a brief description of the model. Afterwards, the critical
parameters of the model, to be used in the forthcoming UQ analysis,
will be identified. Finally, the CFD stagnation-line tool in which the
ablative boundary condition is implemented will be briefly
introduced.
2.1. Theoretical modeling of surface ablation
The N species conservation equations, written in the control
volume limited to the thin lamina representing the gas-surface
interface, take the form of a surface balance [24, 42]:


ðri yÞw þ ri ydi ¼ u_ iw for i ¼ 1; …; N; (1)
w
where ri is the density of the ith
species in the mixture, is the vdi
diffusion velocity, u_ iw is the chemical surface source term, and v is
the bulk velocity of the mixture normal to the surface. The sum of
the advective (ri v) and the diffusive (ri ydi ) terms equates the
source term due to all the surface reactions (Nr ) involving the ith
P r r
species (u_ iw ¼ N
r u
_ iw ). Without losing any generality, one can as-
sume that the surface heterogeneous reactions are the only
responsible for the mass injections. Then, summing up Eq. (1) over
all the species in the mixture, the following global surface mass
balance is obtained:
ðr vÞw ¼ m;
_ (2)
P 2.2. Ablation model input uncertainties characterization
where the term m_ ¼ N i u
_ iw is the mass loss flux of the ablated
material. The source terms of Eq. (1) are defined by the number of
surface reactions included in the analysis. For the present study,
two oxidation reactions (“oxi1 ”: Cs þ O/CO, and “oxi2 ”:
2Cs þ O2 /2CO), one nitridation reaction (“nitri”: Cs þ N/CN),
and one sublimation reaction (“subl”: 3Cs #C3 ) are considered.
Additionally, the surface recombination of atomic nitrogen
(N þ N/N2 ) is also included [43]. Note that usually the reverse
reactions do not need to be considered because of the very low
value of the equilibrium constants [41]. The only exception is the
sublimation reaction for which the reverse process (carbon
condensation) must be allowed. This is done by including the
equilibrium concentration of C3 evaluated through the saturated
carbon vapor pressure (pC;eq: ). The value
pC;eq: ¼ 5:19  1015 expð90845=Tw Þ, taken from Ref. [25], is used
in the model.
The number of the ith gaseous species changed by a surface re-
action per unit time and per unit surface is evaluated by multi-
plying the number flux of the reactant species arriving at the
surface, given by the kinetic theory, by the probability of each re-
action to occur [41]:
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gr 8 R Tw
kri ¼ ; (3)
4 pM i
where kri is the surface reaction velocity, gr the reaction probability,
R the universal gas constant, Tw the surface temperature, and M i
the molar mass of the ith species. This reaction velocity is related to
the surface source term through the species densities: u_ riw ¼ kri ri .
The mass of solid carbon per unit time and per unit surface (mass
blowing flux) removed by each of the heterogenous reactions can
be evaluated as:
499
X
N
m_ j ¼ u_ jiw for j ¼ 1; …; Nj ; (4)
i
being Nj the subset of the Nr surface reactions which involve the
solid carbon. Finally, the total mass blowing flux is computed as:
m_ ¼ m_ oxi1 þ m_ oxi2 þ m_ nitri þ m_ subl : (5)
The thin lamina approach can be used also to formulate an en-
ergy balance over the ablative surface. Considering that all the
energy fluxes arriving at the surface from the flow side are dissi-
pated by the heat conduction through the solid material, the bal-
ance reads:
 N

X 
vT 
lw   hi ri ydi _ s ¼ ðh r vÞw þ q_ cond;s ;
þ q_ radnet þ mh
vr w w
i
(6)
where lw is the gas thermal conductivity, hi the enthalpy of species
i, h the mixture enthalpy, and r indicates the direction normal to the
surface. The subscripts w and s label the gas and the solid sides of
the interface, respectively. Equation (6) includes, from left to right,
the contribution of conduction, diffusion, net radiation (difference
between the incoming and the outgoing radiative fluxes), surface
recession, gas blowing, and solid conduction. This last term can be
approximated using the so-called steady-state ablation hypothesis
[12], becoming a function of mass blowing flux and material


enthalpy: m_ hTs w  hTs b .1 Therefore, Eq. (6) can be solved, together
with Eq. (1), to compute the steady-state surface temperature Tw .
For the evaluation of the of ith species surface source term u_ iw in
Eq. (1), the knowledge of the source terms u_ riw , associated to each of
the surface reactions that the species undergoes, is necessary. The
evaluation of the u_ riw requires the computation of the surface re-
action velocities, which in turn need the knowledge of the surface
reaction probabilities (see Eq. (3)). Reaction probability data can be
differentiated in two main categories: temperature dependent
probabilities and constant probabilities. Probabilities of the first
category are usually expressed by laws that fit the experimental
data with an expression of the form gr ¼ ar expðf ðTÞÞ. The reaction
probability values have been identified as the primary source of
uncertainty for the present ablation model.
We presented the four heterogeneous surface reactions and the
surface recombination reaction included in the present model in
section 2.1. Based on the selected literature data, the oxidation re-
action Cs þ O/CO is the only one featuring a variable reaction
probability among the five considered surface reactions. The
selected nominal values of the probabilities associated to the
considered reactions are listed in Table 1. Literature sources are also
indicated.
The evaluation of the input uncertainties impact on the ablation
model output is the focus of the following analysis. Therefore, a
criterion for the determination of the input quantities has first to be
defined. It is worth noting that behind the definition of a constant
reaction probability often conceals a large scatter of the available
experimental data which complicates the determination of a
1
The steady-state ablation hypothesis assumes i) planar surface; ii) constant
recession velocity; iii) back surface of the material, b, at the initial (cold) temper-
ature, and solid thermal conductivity constant across the material.
500 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509

Table 1 Table 2
Surface reaction probability data and references. Ablation model input uncertainties distributions and ranges along with the related
references used for their definitions.
Reaction Nominal gr Ref.
No. Uncertain parameter Distribution Range Ref.
Cs þ O/CO 0:63expð1160=Tw Þ [41]
2Cs þ O2 /2CO 0.01 [44] 1) aCs þO/CO Uniform 3.70  101 to 1.00  101 [47]
Cs þ N/CN 0.3 [45] 2) g2Cs þO2 /2CO LogUniform 1.00  105 to 1.00  101 [47]
3Cs #C3 1.0 [25] 3) gCs þN/CN Uniform 0. to 3.00  101 [48]
N þ N/N2 0.05 [44]
4) g3Cs #C3 LogUniform 1.00  102 to 1.00 [25]
5) gNþN/N2 Uniform 0. to 5.00  101 [44]
6) εs Uniform 8.00  101 to 9.50  101 [49]

satisfying fitting function. An example of this is given in Fig. 1(a)

where the collected data for the surface nitridation, considered
constant in the model, are shown; the data for atomic-oxygen
oxidation, including the fitting function given in Table 1, are
shown in Fig. 1(b) for comparison.
Therefore, the adopted reasoning for the selection of the un-
certainty distributions is summarized as follows: i) uniform dis-
tribution for the constant gr whose uncertainty range is limited to a
single order of magnitude; ii) log-uniform distribution for the
constant gr whose uncertainty range cover several orders of
magnitude; iii) uniform distribution of the pre-exponential factor
ar for the gr expressed with an exponential law. Table 2 lists the
selected uncertainty distributions together with the ranges and the
reference sources used for their determination. For the reaction
Cs þ O/CO, the rationale behind the given range for the pre-
exponential factor variation was to cover an as large as possible
part of the experimental data (the scanned reaction probabilities lie
between the dotted lines in Fig. 1(b).
Examination of the terms in the surface energy balance, Eq. (6),
points out that the net radiative heat flux should consider the
contributions of both the incoming and the outgoing radiation.
For an accurate estimation of the incoming flux ad-hoc numerical
tools are required [50]. However, it is worth noting that it can
usually be neglected unless extreme conditions are considered
(e.g., very high pressure and/or very high temperature of the gas).
Because of the high emissivity value of carbon based TPS mate-
rials, the outgoing radiation can represent a significant contribu-
tion to the surface energy balance, instead. A reasonable estimate
of this can be computed as q_ radout ¼ sεs Tw
4 (where s is the Stefan-
Boltzmann constant and εs the integral emissivity of the solid
surface). Unfortunately, one has to cope with the fact that an ac-
curate knowledge of the εs value is not always possible. For this
reason, beside the reaction probabilities of the considered surface
heterogeneous reactions, the surface emissivity was included as
an additional uncertain input for the ablation model. The selected
distribution and range for this uncertain parameter are also given
in Table 2. The nominal value used in the model for the surface
emissivity is 0.9 [49].
2.3. Numerical modeling of flow field with ablative boundary
The surface ablation model described in section 2.1 has been
implemented in a stagnation-line code developed at the von Kar-
man Institute (VKI) [33, 51].
This code solves the dimensionally-reduced form of the
Navier-Stokes equations presented in Ref. [52], describing the
reacting flow along the stagnation line of a spherical body. The
spatial discretization of the conservation equations is performed
by means of a cell-centered finite volume method. For the present
study, the Roe's approximate Riemann solver implemented in the
code is used. The diffusive fluxes are evaluated in terms of prim-
itive variables, and the value of these primitive variables and their
gradients at the volume interface are computed by a weighted
average and a central finite difference approximation, respec-
tively. Interested readers are referred to [51, 53] for a detailed
description of this tool.
Imposing the surface balances (Eq. (1) and Eq. (6)) at the
boundary of the computational domain (ablative surface), the
surface state is computed by solving simultaneously the system of
these N þ 1 equations by means of a Newton’s loop; then, the bulk
(blowing) velocity is obtained from Eq. (2). This allows for the direct
computation of the fluxes to be imposed at surface interface when
solving for the state of the physical volume adjacent to the surface.
It is worth to mention that with this approach the ablative

Fig. 1. Examples of surface reaction probability data for two of the considered surface heterogeneous reactions. (a) Cs þ N / CN (from [46]). (b) Cs þ O / CO (from [47]).
 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509 501

boundary is fully coupled with the flow field computation, and the can be represented by a Ndimensional function:
effect of ablation is directly included in mass, momentum, and
energy exchanges between the gas and the surface. y ¼ fðxÞ ¼ fðx1 ; x2 ; /; xN Þ: (8)
The stagnation-line code is coupled with the Mutationþþ li-
brary, also developed at VKI [54]. The library is used for the We consider Eq. (8) in its functional expansion form as follows
computation of the thermodynamic and transport properties, as
well as for the evaluation of the gas-phase chemical source terms. X
N X
N  
Species thermodynamic properties for the present analysis are y ¼ f0 þ fi ðxi Þ þ fij xi ; xj þ /
obtained from the NASA polynomials [55], and relative mixture 1iN 1i < jN
quantities are derived from pure species quantities through mixing þ f1;2;/;N ðx1 ; x2 ; /; xN Þ;
rules. The transport properties are derived from kinetic theory,
which provides relationships for macroscopic transport coefficients or in compact form using a multi index system:
based on microscopic collision integrals.
The library computes the diffusive fluxes based on these data
2X
N
1

and by solving the Stefan-Maxwell equations (thermal and pres- y ¼ fs0 þ fsj xsj ; (9)
sure diffusion are neglected in the present analysis). The chemical j¼1
production rates for species, based on elementary chemical re-
actions including third body, are calculated by taking the forward The multi indices sj are defined such as
reaction rate coefficients specified by the user in an Arrhenius law
form. The backward rate coefficient is determined by satisfying the s0 ¼ ð0; 0; 0; /; 0Þ
equilibrium relation. The library has been designed, implemented s1 ¼ ð1; 0; 0; /; 0Þ
and extensively tested to ensure high fidelity together with low s2 ¼ ð0; 1; 0; /; 0Þ
computational costs. «
sN ¼ ð0; 0; 0; /; 1Þ (10)
3. Uncertainty analysis theory and tools sNþ1 ¼ ð1; 1; 0; /; 0Þ
sNþ2 ¼ ð1; 0; 1; /; 0Þ
Let us consider a stochastic differential equation of the form: «
sN ¼ ð1; 1; 1; /; 1Þ;
Lðx; x; fÞ ¼ f ðx; xÞ; (7)
where N ¼ 2N  1. The representation of Eq. (9) is called ANOVA
where L is a non-linear spatial differential operator (for instance, L (Analysis Of Variance) decomposition [58] of fðxÞ, if for any
is the steady Navier-Stokes operator) depending on a set of un- j2f1; …; N g,
certainties, defined with a random vector x (whose dimension de-
Z
 n o
pends on the number of uncertain parameters in the problem), and
f ðx; xÞ is a source term depending on the position vector x and on x.
fsj xsj pðxi Þdxi ¼ 0 for xi 2 xsj : (11)
The solution of the stochastic Eq. (7) is the unknown dependent ℝ
variable fðx; xÞ, and is a function of the space variable x2Rd and of It follows from Eq. (11) the orthogonality of ANOVA component
x. One of the objective of Uncertainty Quantification is to compute terms, namely
the statistics of the quantity of interest, i.e., fðx; xÞ with respect to
the system uncertainties x.

Using a non-intrusive uncertainty quantification tool means E fsj fsk ¼ 0 for jsk: (12)
that a single deterministic computation (used to solve for example
the differential operator L) is replaced with a whole set of such ANOVA allows identifying the contribution of a given stochastic
computations, each one of which is run for specific values of the parameter to the total variance of an output quantity. Meanwhile,
uncertain conditions. In this work, two non-intrusive techniques we obviously have
are used alternatively to propagate physical uncertainties through

the system under consideration. A polynomial-chaos-based E fsj ¼ 0 for j ¼ 1; …; N :
approach is applied when the problem features a limited number
of uncertainties; this, relying on the polynomial-chaos properties, Note that the terms in the ANOVA decomposition can be
allows an accurate estimation of the statistics for each QoI. An expressed as integrals of fðxÞ. Indeed, we have
anchored-ANOVA approach is applied when larger set of un-
certainties are analyzed, or when some additional parametrical E ðY Þ ¼ f0
analyses of the results obtained with the polynomial-chaos-based
 xi Þ 
EðYj ¼ f 0 þ f i ðxi Þ     (13)
approach are performed. This analysis permits to detect the hier- E Y x i ; x j ¼ f0 þ fi ðxi Þ þ fj xj þ fij xi ; xj ;
archy of the most predominant uncertainties and significantly
reduced the number of deterministic simulations needed to and so on, where EðYj,Þ denotes the conditional expectation with
perform the stochastic analysis. respect to the conditional PDF defined in the standard way.
Both methods, starting from the second for ease of notation, are In order to introduce the less expensive anchored ANOVA, the
described briefly in the following. For more details, refer to [56] for Dirac measure is used in integrations in Eq. (13):
the Polynomial-based method and to [57] for the anchored-ANOVA
approach.
pðxi Þdxi ¼ dðxi  ci Þdxi for i ¼ 1; …; N: (14)
3.1. Anchored-ANOVA approach: definitions and basic notions Thus, pðxÞdx ¼ dðx  cÞdx. The point c ¼ ðc1 ; /; cN Þ is called
“anchor point.” Hence, the ANOVA component terms in Eq. (13) can
Let us suppose that the response of a given system of interest be expressed as follows:
502 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509

fðcÞ ¼ f0 hfðx; xÞ; Ji ðxÞi

 xi Þ 
fðcj ¼ f0 þ fi ðxi Þ    
f c x i ; x j ¼ f0 þ fi ðxi Þ þ fj xj þ fij xi ; xj
«:
The formulae in Eq. (15) are used to quantify the expectation
and variance of the component functions, by simply evaluating
the model output at chosen sampling points. For more details,
see Refs. [57, 59]. This permits a strong reduction of the
computational cost, since this avoids the computation of several
integrals. Moreover, a variance-based adaptive criterion (see for
more details [59]) is used in order to compute the so-called
effective dimension and to evaluate high-order interactions
with a reduced computational cost. The order at which the
ANOVA model is truncated, called effective dimension, is the one
beyond which the difference between the ANOVA model and the
truncated expansion in a certain measure is very small. Given a
threshold, this implies that the smallest terms representing those
interactions which contribute to exceed that threshold are
dropped.
In this paper, a covariance decomposition of the output variance
has been considered, as proposed in Ref. [57], in order to accurately
compute the statistics using the anchored-ANOVA expansion. The
covariance decomposition makes the result less sensitive to the
choice of the anchor point if a full expansion of the anchored
ANOVA is employed (the reader is referred to [60, 61] for the
sensitivity problem of classical anchored-ANOVA method con-
cerning the choice of the anchor point).
Note that in this work we always refer to the so-called Total
Sensitivity Index (TSI), since the aim is to compute the overall in-
fluence of a variable. This sensitivity index can be computed sum-
ming up all the contributions in which the chosen variable is
present.
3.2. Polynomial-chaos-based approach: definitions and basic
notions
Under specific conditions, a stochastic process can be expressed
as a spectral expansion based on suitable orthogonal polynomials,
with weights associated to a particular probability density function.
The first study in this field is the Wiener (1938) process. The basic
idea is to project the variables of the problem onto a stochastic
space spanned by a complete set of orthogonal polynomials J that
are functions of random variables x. For example, the unknown
variable f has the following spectral representation:
X
∞ 3 MW/m2 cold-wall heat flux. Table 3 lists the free-stream condi-
fðx; xÞ ¼ fi ðxÞJi ðxÞ:
i¼0
In practice, the series in Eq. (16) has to be truncated to a finite
number of terms, here denoted with N. The total number of terms
of the series is determined by:
ðn þ p0 Þ !
Nþ1¼ ; (17)
n!p0 !
where n is the dimensionality of the uncertainty vector x and p0 is
the order of the polynomial expansion. Each polynomial Ji ðxÞ of
total degree no is a multivariate polynomial form, which involves
tensorization of the 1D polynomial form. The polynomial basis is
chosen accordingly to the Wiener-Askey scheme [62] in order to
fi ¼ ; ci: (18)
(15) hJi ðxÞ; Ji ðxÞi
Several approaches can be used to estimate PC coefficients. The
approach selected in this study is based on quadrature formula. As a
consequence, we obtain the solution of a deterministic problem for
each quadrature point (see Ref. [56] for further details). In both
cases, once the chaos polynomials and the associated fi coefficients
are computed, the mean and the variance of the stochastic solution
fi ðx; xÞ are obtained from:
EPC ¼ f0 ðxÞ; (19)
* +
X
N
VarPC ¼ f2i ðxÞ J2i : (20)
i¼1
Another appreciable feature of PC expansion is that it simplifies
the sensitivity analysis based on the ANOVA decomposition, which
can be easily performed by using some interesting properties of the
previous development [63]. In fact, the contribution to the variance
of a given random variables with index k, i.e., the first order Sobol'
index, can be obtained as follows:
P D E
2
i2a fi J2i ðxÞ
Sk ¼ ; (21)
VarPC
where a represent the set of indexes associated to a given uncer-
tainty k (we strongly recommended [63] for more details). Note
that for problems featuring a large number of uncertainties, the
computational cost could become prohibitive and, even if some
cures exist, e.g., the use of Sparse-Grid approaches, the cost can still
be unfeasible. Mainly for this reason, we have included in this work
the lower-cost Anchored-ANOVA method presented in the previous
section.
4. Uncertainty analysis on ablation modeling
In the following, the deterministic CFD tool described in section
2 and the stochastic tools presented in section 3 are used to
perform the uncertainty analysis. The uncertain parameters are the
variables given in Table 2. We selected as numerical test case a
high-enthalpy air flow around a graphite sphere (2.5 cm radius).
The test conditions are representative of an ablation experiment
that could be performed in the inductively-coupled plasma torch
“Plasmatron” facility at VKI: 16 g/s mass flow rate at 200 hPa with a
(16) tions for the selected test case. A mixture of 15 species (C, C2 , C3 , CN,
CO, CO2 , N, N2 , NO, O, O2 , Cþ , Nþ , Oþ , e ) was used in the simu-
lations. The gas-phase reaction mechanisms were selected from a
subset of those given in Ref. [64].
As described in the introduction, ablative TPS materials are used
in a wide range of aerospace applications. Therefore, the definition
of the QoIs can vary among the different applications. In our study
we focus on plasma wind tunnel tests for spacecraft heat-shield
material, and the followings were identified as possible
Table 3
Nominal inlet conditions for the analyzed test case.
Variable Value

select orthogonal polynomials with respect to the probability Velocity, m/s 53

density function pðxÞ of the input. The orthogonality can be ad- Density, kg=m3 4:54  103
vantageously used to compute the coefficients of the expansion in a Temperature, K 7818
Mach 2:1  102
non-intrusive PC framework
 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
phenomena of interest: i) surface regression; ii); material thermal
response. These were eventually tracked down to two QoIs, namely
the mass blowing flux m, _ and the surface temperature Tw .
In this section we first present the nominal output of the
deterministic simulation with the ablative boundary condition.
Afterwards, the uncertainty analysis, performed with the PC
method introduced in section 3.2, is described. Finally, a corollary
study on the effect of the uncertain input distributions is carried out
using an anchored-ANOVA approach (section 3.1). For each of these
analyses two scenarios are investigated: i) the full set of uncertain
input quantities presented in Table 2 is used; ii) surface nitridation
is inhibited and only the remaining five uncertainties are consid-
ered. For ease of reading, these two cases will be also referred to as
scenario A and scenario B hereafter.
4.1. Ablation model nominal output
Figure 2 shows the boundary-layer profiles of the major species
and the boundary-layer temperature distributions for the two
scenarios. The main product of the surface reactions is CO, followed
by CN and C3 (Fig. 2(a)). Figure 2(b) shows that a higher steady-
state surface temperature is obtained from the energy balance,
Eq. (6), if nitridation is neglected. This behavior is a first indication
of a possible coupling between surface nitridation and surface
recombination of N in N2 . Figure 2(a) shows totally different CN
concentration profiles for the two scenarios. Including nitridation
causes a strong reduction of the nitrogen concentration at the
surface. Therefore, less nitrogen is available to undergo the
recombination as well. This explains the reduced steady-state
temperature for scenario A. The higher surface temperature for
scenario B also triggers the surface sublimation, increasing the
production of C3 .
The concentration of the ablation products strongly decreases
through the boundary layer and is almost negligible at its edge
(about 5 mm from the surface), where the relevant species are
roughly those of air. A very low contribution of the diatomic-oxygen
oxidation to the mass blowing flux is foreseen if the reactant con-
centration through the boundary layer is analyzed. In fact, the O2
concentration is already very low at the boundary layer edge (ox-
ygen almost fully dissociates in the present conditions) and no
significant recombination takes place through the boundary layer.
Table 4 lists the QoIs values obtained for the nominal cases of
the two scenarios. The effect of nitridation on the surface mass
blowing flux is evident. This is a consequence of the high value
Fig. 2. Stagnation-line boundary-layer profiles of mass fractions (a) and temperature (b) obtained from the nominal deterministic simulations with the ablative boundary condition.
Both scenarios, including and neglecting the surface nitridation, are shown. (a) Relevant mass fractions. (b) Temperature.
503
Table 4
QOIs obtained from the nominal deterministic stagnation-line simulation with the
ablative boundary condition.
QoI w/ nitridation w/o nitridation
_ kg/(m2s)
m, 4:08  102 2:11  102
Tw , K 2534 2840
selected for the nitridation reaction probability (Table 1). Note that,
although very low values of atomic-oxygen mass fractions are
found for both scenarios at the gas-surface interface (see Fig. 2(a)),
the presence of nitridation strongly decreases the contribution of
atomic-oxygen oxidation to the total mass blowing flux
(m_ oxi1 ¼ 2:11  102 w/o nitridation and m_ oxi1 ¼ 7:55  103 w/
nitridation). This suggests that, for the present conditions, the
oxidation process is limited by diffusion rather than by the reaction
kinetics.
The surface temperature is higher for scenario B (see Table 4).
This is ascribed to both atomic-oxygen oxidation and surface ni-
trogen recombination. Removing nitridation, in fact, has only a
marginal effect on the surface energy balance because the reaction
is slightly exothermic (0.34eV are released). Oxidation instead,
exothermic with a release of energy of 3.74eV, is more prominent in
the scenario B. Moreover, as already pointed out, omitting nitrida-
tion indirectly affects nitrogen recombination making more nitro-
gen available to the exothermic catalytic process. Overall, this
results in an increase of the energy transferred to the surface.
4.2. Uncertainty analysis
We first analyze the results of the uncertainty analysis per-
formed with the PC method described in section 3.2. As for the
nominal simulations, scenario A and scenario B are both analyzed.
Relying on classical quadrature formulae for uniform PDF (see
Ref. [63] for more details), the deterministic simulations necessary
to build up the 2nd order sample plans were 729 (6 uncertain input
quantities) and 243 (5 uncertain input quantities) for scenario A and
scenario B, respectively.
The computed mean, variance and the percent coefficient of
variation (e% ) are given in Table 5. The mean values of mass blowing
flux are different for the two scenarios. The difference between
them is significant, even though it is lower than what found for the
nominal cases (Table 4). Moreover, the higher variance of m_ found
for scenario A suggests that a direct effect of nitridation is the main
504 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509

Table 5 strong coupling between nitrogen recombination and carbon

QOI values obtained from the 2nd order polynomial-chaos UQ analysis using the nitridation is envisaged, and one can conclude that the uncertainty
input distributions of Table 2.
on the surface nitridation reaction probability is the main respon-
QoI Mean Variance e% sible for the high percent coefficient of variation (e% ) given in
w/ nitridation _ kg/(m2s)
m, 3.10  102 2.69  105 ±16:72 Table 5 for scenario A.
Tw , K 2722 15400 ±4:56 In contrast, for scenario B, the variance of the mass blowing flux
w/o nitridation _ kg/(m2s)
m, 2.10  102 2.63  1010 ±1:15 is significantly lower. From Fig. 3(b) it can be seen that the domi-
Tw , K 2903 2740 ±1:80 nant contribution to this variance comes from atomic-oxygen
oxidation. In this case, the effect of recombination is lower, con-
firming that the coupling with nitridation was the main reason for
responsible for the discrepancies between the two scenarios. This is the high influence of the recombination on m_ found in the previous
confirmed if an ANOVA analysis is performed, and the so-called scenario. It is interesting to note that the computed e% for scenario B
Sobol’ indexes are computed to show the influence of each of the is rather small although the non-negligible uncertain range
uncertain input quantities on the QoIs. resulting for the atomic-oxygen oxidation reaction probability (e.g.,
Figure 3 shows the computed ANOVA contributions of the mass ranging between 0.23 and 0.63 for a wall temperature of 2530 K).
blowing flux for the two scenarios. Both surface nitrogen recom- This somehow supports the analysis presented for the nominal
bination and surface nitridation seem to contribute strongly to the results, where the diffusion limited regime of this reaction was
variance for scenario A (Fig. 3(a)). However, a direct effect of the supposed for the present conditions. Moreover, the Sobol’ index for
recombination on the mass blowing flux is unlikely for the recombination is comparable to the wall emissivity one, suggesting
following reasons: i) the recombination is “passive” and does not that they both affect the m_ variance by modifying the surface
consume the material; ii) the primary effect of the recombination is temperature. Therefore, being reasonable to consider the oxidation
to increase the surface temperature, but the computed Sobol’ index to be diffusion limited for the reasons given above, the influence of
for the atomic-oxygen oxidation reaction, which is the only one the recombination and the emissivity can be explained through the
having a reaction probability dependent on temperature, is very boundary-layer temperature effect on the diffusion fluxes.
low (even though in section 4.1 it has been shown that it contrib- Figure 4 confirms that the main contribution to the surface
utes substantially to the global mass blowing flux). Therefore, a temperature variance is coming from the surface nitrogen

_ Sobol’ indexes obtained from the 2nd order polynomial-chaos UQ analysis using the input distributions of Table 2. (a) w/ nitridation. (b) w/o
Fig. 3. Mass blowing flux (m)
nitridation.

Fig. 4. Surface temperature (Tw ) Sobol’ indexes obtained from the 2nd order polynomial-chaos UQ analysis using the input distributions of Table 2. (a) w/ nitridation. (b) w/o
nitridation.
 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
recombination. Conclusions similar to those deducted for the mass
blowing flux can be drawn comparing the two scenarios. The en-
ergy released from the surface nitridation, 0.34 eV, is fairly low.
Therefore, the high influence of nitridation on the surface tem-
perature of scenario A, most probably comes from the coupling with
nitrogen recombination. The high uncertainty in the nitridation
reaction probability for scenario A causes the temperature e% to be
almost three times higher than the one of scenario B. For this second
scenario, the influence of nitrogen recombination and surface
emissivity are comparable.
Overall, the uncertainty analysis seems to suggest that the un-
certainty on the nitridation reaction probability has the biggest
influence on the present ablation model in the selected conditions.
However, one can argue that the results are prone to the somehow
arbitrary selection of the uncertain input distributions. The accu-
rate choice of the distribution for the uncertain input parameters is
indeed one of the key parameter in every uncertainty analysis. The
influence of the input distribution is therefore analyzed in the
following.
4.3. Effect of input distribution
As already pointed out, the uncertainty analysis presented in
section 4.2 required a non-negligible number of deterministic
simulations in order to build up the sample plan: 729 simulations
for scenario A and 243 simulations for scenario B. Although the
deterministic stagnation-line code allows fast computations, the
time needed for the generation of the sample plans was not
negligible. The anchored-ANOVA method described in section 3.1
has the advantage of reducing considerably the number of deter-
ministic computations, and was therefore considered suitable for
this analysis on the effect of the input distributions.
First we wanted to verify the accuracy of the method with
respect to the PC expansion used in the previous uncertainty
analysis. We used the input distributions of Table 2 and we
analyzed again both scenarios. The analysis was divided in the
following steps: i) three different 1st order analyses per scenario
were performed by varying the number of samples of each
dimension (4, 8, 16 samples per dimension were used); ii) a 2nd
order analysis, which uses the variance-based adaptive criterion
mentioned in Section 3.1, was carried out based on the results of the
most reliable 1st order analysis.
For the first step of scenario A, mean and variance of both QoIs
resulted almost unchanged with respect to the previous analysis.
Moreover, the Sobol’ indexes changed imperceptibly for the three
_ total sensitivity indexes (TSIs) obtained from the 2nd order anchored-ANOVA UQ analysis using the input distributions of Table 2. (a) w/ nitridation.
Fig. 5. Mass blowing flux (m)
(b) w/o nitridation.
505
different sample numbers (24, 48, and 96 samples). In the case of
scenario B, a small variation of e% , between 2.12% and 2.68%, was
observed on the surface temperature for the three cases (20, 40,
and 80 samples). On this basis, the results obtained with 4 samples
per dimension were considered satisfactory enough. Therefore,
they have been used afterwards as the base for the 2nd order
analysis. It is worth noting that, from both the analyses presented
so far (the 2nd order PC and the 1st order anchored-ANOVA one)
the number of influential input quantities was found to be limited
for all the scenarios. Therefore, it was not necessary to exploit the
potential advantage of using a “heavy” adaptivity in the anchored-
ANOVA approach in order to further reduce the number of di-
mensions of the 2nd order anchored-ANOVA analyses.
Figures 5 and 6 show the total sensitivity indexes (defined in
section 3.1) as obtained from the 2nd order analyses for the two
scenarios. As seen they compare well with the results shown in
section 4.2. Moreover, the obtained results are very similar to the
corresponding 1st order analysis. For scenario A, the deviation on e%
is of 1:13% for the mass blowing and 0:3% for the surface tem-
perature; for scenario B it is < 0:1% for the mass blowing flux and
0:1% for the surface temperature. Overall this analysis proved that
the anchored-ANOVA approach results were in good agreement
with those of the PC analysis for the same input distributions.
To understand how the results of the analysis presented in
section 4.2 would be affected if a different input distribution was
selected, an analysis with the anchored-ANOVA method was per-
formed considering a LogUniform distribution for all the consid-
ered uncertainties. The previous results showed that the influential
uncertainties are only a subset of those given in Table 2. Therefore,
only the relevant uncertain input quantities were considered in the
present case (i.e., number 1, 3, 5, and 6 in Table 2). Table 6 lists the
selected input quantities and their distributions for the two sce-
narios (the ranges are assumed equal to those in Table 2).
The results of the 2nd order analyses are given in Table 7 for the
two scenarios. First of all, an evident homogenization between the
results of the two scenarios is found. The mass blowing flux of
scenario A is noticeably reduced with respect to the analysis shown
in section 4.2, whereas the one of scenario B still agrees with the
previous results. A cause of the reduction found for scenario A is
undoubtedly the fact that with the LogUniform distribution the
surface nitridation reaction probability encompasses very low
ranges, not covered when the uniform distribution was used.
However, the strong coupling between nitrogen recombination and
carbon nitridation is still evident from the TSIs of the mass blowing
flux in scenario A (Fig. 7(a)). Note that, with respect to the results in
506 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509

Table 6 Different results are obtained for the surface temperature, that is
Ablation model input uncertainties and modified distributions. significantly different from the results presented in section 4.2 for
Uncertain parameter Distribution both scenarios. In particular, the drop in the mean value for scenario
B is close to 350 K. This is, most probably, the consequence of a
w/ nitridation aCs þO/CO LogUniform
gCs þN/CN LogUniform better scan of the low values of the nitrogen recombination prob-
gNþN/N2 LogUniform ability range obtained with the distribution of Table 6. The bigger
εs LogUniform influence of nitrogen recombination is the cause for the noticeable
growth of e% for scenario B in Table 7. The two scenarios present
w/o nitridation aCs þO/CO LogUniform
gNþN/N2 LogUniform now similar e% for the surface temperature. The explanation for this
εs LogUniform result can be the reduced “effective” coupling between nitridation
and recombination in scenario A. In fact, if lower average values of
the nitridation probability are covered, it is reasonable to imagine a
Table 7
limited variation of the nitrogen surface concentration caused by
QOI values obtained from the 2nd order anchored-ANOVA UQ analysis using the the uncertainty on the nitridation probability. The differences with
modified input distributions of Table 6. respect to the previous analysis are evident if Fig. 8 is compared to
QoI Mean Variance e% Fig. 4. Now the surface recombination is by far the most influential
parameter on the surface temperature for both scenarios.
w/ nitridation _ kg/(m2s)
m, 2.52  102 4.55  105 ±26:78
Tw , K 2544 26178 ±6:36

w/o nitridation _ kg/(m2s)

m, 2.10  102 7.10  108 ±1:27 4.4. Discussion
Tw , K 2558 28160 ±6:56
We performed several analyses. Our outcomes suggest that, for
the analyzed conditions, the ablation model is quite sensitive to
two main uncertainties: the carbon nitridation probability and the
Table 5, the e% of this QoI is increased for scenario A and almost
nitrogen recombination one. However, as shown, the arbitrary
unchanged for scenario B. An overall look at the TSIs in Fig. 7 sug-
choice on the input distributions can have a strong influence on the
gests a very good agreement with the Sobol' indexes given in Fig. 3.
results of such kind of uncertainty analysis. This has to be kept in

Fig. 6. Surface temperature (Tw ) total sensitivity indexes (TSIs) obtained from the 2nd order anchored-ANOVA UQ analysis using the input distributions of Table 2. (a) w/ nitridation.
(b) w/o nitridation.

Fig. 7. Mass blowing flux (m) _ total sensitivity indexes (TSIs) obtained from the 2nd order anchored-ANOVA UQ analysis using the modified input distributions of Table 6. (a) w/
nitridation. (b) w/o nitridation.
 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
Fig. 8. Surface temperature (Tw ) total sensitivity indexes (TSIs) obtained from the 2nd order anchored-ANOVA UQ analysis using the modified input distributions of Table 6. (a) w/
nitridation. (b) w/o nitridation.
mind when quantitative error margins of the numerical results are
sought through this stochastic approach. We remark the fact that a
high uncertainty of these key parameters was selected on purpose
in the present numerical exercise, where the goal was to analyze
one of the worst possible scenarios.2 If more reasonable ranges for
these parameters are selected, the uncertainty analysis results will
certainly benefit from it.
We finally note that this type of analysis is much more general
than a local sensitivity analysis, i.e., the computation of a gradient
with respect to some input parameters. In fact, the influence of the
probability density function of the input parameters is here taken
into account. Moreover, the presented analysis is valid not only
around the nominal condition, which is the case for a local sensi-
tivity analysis, but in the whole space of variability of the input
parameters.
5. Conclusions
Separately developed deterministic and stochastic tools have
been used together in the study. A stagnation-line code featuring an
ablative boundary condition has served as the deterministic tool;
two different numerical codes have been used independently as the
stochastic tools, depending on the circumstances. These two nu-
merical codes are based on two different uncertainty quantification
approaches, the polynomial-chaos expansion method and the
anchored-ANOVA one. However, the analysis has shown that they
give equivalent results when applied to the same problem.
The main uncertainties of the CFD ablation model were identi-
fied in the reaction probabilities that drive the surface reactions
included in the model: two oxidations reactions, a nitridation re-
action, a carbon sublimation reaction, and a nitrogen recombina-
tion reaction. The material emissivity was also included as
additional uncertainty in the analysis. Among all the considered
uncertainties, the reaction probability of the surface nitridation was
identified as the most critical one, due to its high variability found
in the literature data. For this reason we studied two separate
scenarios: including (with a defined uncertainty) and neglecting
surface nitridation.
The studied test case was the forward-stagnation-point ablation
of a non-pyrolyzing (e.g., graphite) sphere of 2.50 cm radius
immersed in a high-enthalpy flow. Selected flow conditions were
2
For instance, the upper bound value of the nitridation reaction probability,
taken from Ref. [45], seems to be inexplicably high with respect to more recent
references [48,65e67].
507
similar to those of a typical inductively coupled plasma wind tunnel
tests (200 hPa of static pressure, Mach number of 0.02, and 3 MW of
cold-wall heat flux). Ablation mass blowing flux and steady-state
surface temperature were identified as the quantities of interests
(QoIs) of the study.
The surface nitridation showed a strong influence on the mass
blowing flux in the first scenario (i.e., w/ surface nitridation among
the uncertainties), and two main effects have been identified. The
first is the significant change (17%) of the percent coefficient of
variation for the mass blowing flux. The second is the occurrence of
a strong coupling between nitridation and nitrogen surface
recombination. This coupling causes a non-negligible, unexpected,
influence of nitrogen recombination on the surface mass blowing
flux. Similar conclusions are drawn analyzing the results obtained
for the steady-state surface temperature. Nitrogen recombination
showed the highest impact on this QoI. However, although only
slightly exothermic (0.340 eV), nitridation resulted the second most
influential parameter. This “apparent” strong influence of nitrida-
tion is ascribed to the mentioned coupling with nitrogen
recombination.
A strong reduction of the percent coefficients of variation was
observed for both the mass blowing flux and the surface temper-
ature in the second scenario (i.e., w/o surface nitridation among the
uncertainties). The percent coefficient of variation of the mass
blowing flux dropped to about 1/14 of the value obtained for the
first scenario. The influence of nitrogen recombination on the mass
blowing flux almost disappeared confirming somehow the pres-
ence of the suggested coupling between nitridation and nitrogen
recombination. Moreover, we observed that: i) although the main
contribution to the mass blowing flux comes from atomic-oxygen
oxidation, its reaction probability uncertainty shows a non-
negligible impact on the variance only when nitridation is
omitted; ii) even if an influence of the atomic-oxygen oxidation
uncertainty is found for the second scenario, the computed vari-
ance of the mass blowing flux is noticeably low. We concluded that
the reason for this behavior is the diffusion limited regime of
atomic-oxygen oxidation in the analyzed conditions. Results on the
steady-state temperature for this second scenario showed that ni-
trogen recombination and material emissivity are the two domi-
nant uncertainties.
The CFD modeling of the gas-surface interaction over ablative
thermal protection system materials is largely used for the design
and the analysis of aerospace missions. This analysis can represent
a valuable example of the application of modern uncertainty
quantification methods to this kind of modeling approach.
508 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
However, it has to be kept in mind that the strong influence of a
particular subset of the uncertain input quantities, as found in this
analysis, is strictly connected to the selected test conditions (e.g.,
atomic-oxygen oxidation has been found to be in the diffusion
limited regime for the selected test case). Future, more compre-
hensive, studies should analyze different conditions in order to
draw more general conclusions. Having said this, we believed that
more advanced uses of the CFD ablation model, as the one pre-
sented in Ref. [46] to extract useful model parameters or material
characteristics through a joint numerical-experimental approach,
would certainly benefit from such a kind of uncertainty analysis.
A final remark is related to possible presence of additional un-
certainties, not directly related to the ablation model, which may
have an impact on the numerical rebuilding of a real plasma wind
tunnel experiment (i.e., free-stream condition determination). This
will be the subject of the second part of this work, where the full
procedure used at the von Karman Institute to rebuild the flow
conditions will be studied with an uncertainty analysis to verify the
impact of these additional uncertainties on the final QOIs.
Acknowledgements
The research of A. Turchi and T. Magin is supported by the Eu-
ropean Research Council Starting Grant #259354.
References
[1] Anderson Jr JD. Hypersonic and high-temperature gas dynamics. second ed.
Reston, Virginia: AIAA Education Series; 2006.
[2] Laub B, Venkatapathy E. Thermal protection system technology and facility
needs for demanding future planetary missions. In: Proceedings of interna-
tional workshop on planetary probe atmospheric entry and descent trajectory
analysis and science. Noordwijk, The Netherlands: ESTEC, ESA SP-544; 2004.
p. 239e47.
[3] Scotti SJ. Current technology for thermal protection systems. NASA Confer-
ence Publication CP-3157; 1992.
[4] Duffa G. Ablative thermal protection systems modeling. Reston, Virginia: AIAA
Education Series; 2013.
[5] Park C. Effects of atomic oxygen on graphite ablation. AIAA J 1976;14:1640e2.
http://dx.doi.org/10.2514/3.7267.
[6] Allendorf H, Rosner D. Comparative studies of the attack of pyrolytic and
isotropic graphite by atomic and molecular oxygen at high temperatures.
AIAA J 1968;6:650e4. http://dx.doi.org/10.2514/3.4558.
[7] Jacobson NS, Curry DM. Oxidation microstructure studies of reinforced car-
bon/carbon. Carbon 2006;44:1142e50. http://dx.doi.org/10.1016/
j.carbon.2005.11.013.
[8] Shemet VZ, Pomytkin A, Neshpor V. High-temperature oxidation behaviour of
carbon materials in air. Carbon 1993;31:1e6. http://dx.doi.org/10.1016/0008-
6223(93)90148-4.
[9] Panerai F, Martin A, Mansour NN, Sepka SA, Lachaud J. Flow-tube oxidation
experiments on the carbon preform of a phenolic-impregnated carbon abla-
tor. J Thermophys Heat Tr 2014;28:181e90. http://dx.doi.org/10.2514/
1.T4265.
[10] Helber B, Asma CO, Babou Y, Hubin A, Chazot O, Magin TE. Material response
characterization of a low-density carbon composite ablator in high-enthalpy
plasma flows. J Mater Sci 2014;49:4530e43. http://dx.doi.org/10.1007/
s10853-014-8153-z.
[11] Helber B, Turchi A, Scoggins JB, Hubin A, Magin TE. Experimental investigation
of ablation and pyrolysis processes of carbon-phenolic ablators in atmo-
spheric entry plasmas. Int J Heat Mass Tran 2016;100:810e24. http://
dx.doi.org/10.1016/j.ijheatmasstransfer.2016.04.072.
[12] Milos F, Rasky DJ. Review of numerical procedures for computational surface
thermochemistry. J Thermophys Heat Tr 1994;8:24e34. http://dx.doi.org/
10.2514/3.497.
[13] Duffa G, Vignoles GL, Goyhe ne

che J, Aspa Y. Ablation of carbon-based mate-
rials: investigation of roughness set-up from heterogeneous reactions. Int J
Heat Mass Tran 2005;48:3387e401. http://dx.doi.org/10.1016/
j.ijheatmasstransfer.2005.02.036.
[14] Vignoles GL, Lachaud J, Aspa Y, Goyhe ne
che J. Ablation of carbon-based ma-
terials: multiscale roughness modelling. Compos Sci Technol 2009;69:
1470e7. http://dx.doi.org/10.1016/j.compscitech.2008.09.019.
[15] Nouri N, Martin A. Numerical study of geometrical effects on charring ablative
arc-jet samples. Int J Heat Mass Tran 2015;80:629e35. http://dx.doi.org/
10.1016/j.ijheatmasstransfer.2014.12.041.
[16] Li W, Huang H, Tian Y, Zhao Z. Nonlinear analysis on thermal behavior of
charring materials with surface ablation. Int J Heat Mass Tran 2015;84:
245e52. http://dx.doi.org/10.1016/j.ijheatmasstransfer.2015.01.004.
[17] Lin W-S. Quasi-steady solutions for the ablation of charring materials. Int J
Heat Mass Tran 2007;50:1196e201. http://dx.doi.org/10.1016/
j.ijheatmasstransfer.2006.11.011.
[18] Lachaud J, Aspa Y, Vignoles GL. Analytical modeling of the steady state abla-
tion of a 3D C/C composite. Int J Heat Mass Tran 2008;51:2614e27. http://
dx.doi.org/10.1016/j.ijheatmasstransfer.2008.01.008.
[19] Kendall RM, Barlett EP, Rindal RA, Moyer CB. An analysis of the chemically
reacting boundary layer and charring ablator. Part I: Summary Report, NASA
Contractor Report CR-1060. 1968.
[20] Milos FS, Chen YK. Two-dimensional ablation, thermal response, and sizing
program for pyrolyzing ablators. J Spacecr Rockets 2009;46:1089e99. http://
dx.doi.org/10.2514/1.36575.
[21] Weng H, Bailey SC, Martin A. Numerical study of iso-Q sample geometric ef-
fects on charring ablative materials. Int J Heat Mass Tran 2015;80:570e96.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2014.09.040.
[22] Lachaud J, van Eekelen T, Scoggins JB, Magin TE, Mansour NN. Detailed
chemical equilibrium model for porous ablative materials. Int J Heat Mass
Tran 2015;90:1034e45. http://dx.doi.org/10.1016/j.ijheatmasstransfer.
2005.02.036.
[23] Li W, Huang H, Ai B, Zhang Z. On the novel designs of charring composites for
thermal protection application in reentry vehicles. Appl Therm Eng 2016;93:
849e55. http://dx.doi.org/10.1016/j.applthermaleng.2015.10.070.
[24] Keenan JA. Thermo-chemical ablation of heat shields under earth Re-entry
conditions. Ph.D. thesis. Raleigh, North Carolina: North Carolina State Uni-
versity; 1994.
[25] Chen Y, Milos F. Navier-stokes solutions with finite rate ablation for planetary
mission earth reentries. J Spacecr Rockets 2005;42:197e205. http://
dx.doi.org/10.2514/1.12248.
[26] Bianchi D, Nasuti F, Martelli E. Navier-stokes simulations of hypersonic flows
with coupled graphite ablation. J Spacecr Rockets 2010;47:554e62. http://
dx.doi.org/10.2514/1.47995.
[27] Johnston CO, Gnoffo PA, Mazaheri A. Study of ablation-flowfield coupling
relevant to the Orion heatshield. J Thermophys Heat Tr 2012;26:213e21.
http://dx.doi.org/10.2514/1.T3769.
[28] Bianchi D, Nasuti F, Martelli E. Thermochemical erosion analysis for graphite/
carbon-carbon rocket nozzles. J Propul Power 2011;27:197e205. http://
dx.doi.org/10.2514/1.47754.
[29] Thakre P, Yang V. Chemical erosion of carbon-carbon/graphite nozzles in
solid-propellant rocket motors. J Propul Power 2008;24:822e33. http://
dx.doi.org/10.2514/1.34946.
[30] Turchi A, Bianchi D, Nasuti F, Onofri M. A numerical approach for the study of
the gas-surface interaction in carbon-phenolic solid rocket nozzles. Aerosp Sci
Technol 2012;27:25e31. http://dx.doi.org/10.1016/j.ast.2012.06.003.
[31] Bianchi D, Turchi A, Nasuti F, Onofri M. Chemical erosion of carbon-phenolic
rocket nozzles with finite-rate surface chemistry. J Propul Power 2013;29:
1220e30. http://dx.doi.org/10.2514/1.B34791.
[32] Driver DM, MacLean M. Improved predictions of Pica recession in arc jet shear
tests. AIAA Paper 2011-141. 2011. http://dx.doi.org/10.2514/6.2011-141.
[33] Turchi A, Helber B, Munafo
A, Magin T. Development and testing of an abla-
tion model based on plasma wind tunnel experiments. AIAA Paper 2014-
2125. 2014. http://dx.doi.org/10.2514/6.2014-2125.
[34] Chen Y-K, Go € kçen T. Implicit coupling approach for simulation of charring
carbon ablators. J Spacecr Rockets 2014:1e10. http://dx.doi.org/10.2514/
1.A32753.
[35] Martin A, Boyd ID. Strongly coupled computation of material response and
nonequilibrium flow for hypersonic ablation. J Spacecr Rockets 2015;52:
89e104. http://dx.doi.org/10.2514/1.A32847.
[36] Le Maître O, Knio OM. Spectral methods for uncertainty quantification: with
applications to computational fluid dynamics. Springer Science & Business
Media; 2010.
[37] Xiu D. Numerical methods for stochastic computations: a spectral method
approach. Princeton University Press; 2010.
[38] Mendes M, Ray S, Pereira J, Pereira J, Trimis D. Quantification of uncertainty
propagation due to input parameters for simple heat transfer problems. Int J
Therm Sci 2012;60:94e105. http://dx.doi.org/10.1016/
j.ijthermalsci.2012.04.020.
[39] Bose D, Brown JL, Prabhu DK, Gnoffo P, Johnston CO, Hollis B. Uncertainty
assessment of hypersonic aerothermodynamics prediction capability.
J Spacecr Rockets 2013;50:12e8. http://dx.doi.org/10.2514/1.A32268.
[40] Wright MJ, Bose D, Chen Y-K. Probabilistic modeling of aerothermal and
thermal protection material response uncertainties. AIAA J 2007;45:399e410.
http://dx.doi.org/10.2514/1.26018.
[41] Park C, Jaffe RL, Partridge H. Chemical-kinetic parameters of hyperbolic earth
entry. J Thermophys Heat Tr 2001;15:76e89. http://dx.doi.org/10.2514/
2.6582.
[42] Turchi A. A gas-surface interaction model for the numerical study of rocket
nozzle flows over pyrolyzing ablative materials. Ph.D. thesis. Rome, Italy:
University of Rome “La Sapienza”; 2013.
[43] Driver D, Olson M, Barnhardt M, MacLean M. Understanding high recession
rates of carbon ablators seen in shear tests in an arc jet. AIAA Paper 2010-
1177. 2010. http://dx.doi.org/10.2514/6.2010-1177.
[44] MacLean M, Marschall J, Driver DM. Finite-rate surface chemistry model, II:
coupling to viscous Navier-Stokes code. AIAA Paper 2011-3784. 2011. http://
dx.doi.org/10.2514/6.2011-3784.
 A. Turchi et al. / International Journal of Thermal Sciences 118 (2017) 497e509
[45] Park C, Bogdanoff D. Shock-tube measurement of nitridation coefficient of
solid carbon. J Thermophys Heat Tr 2006;20:487e92. http://dx.doi.org/
10.2514/1.15743.
[46] Helber B, Turchi A, Chazot O, Magin T. Physico-chemistry of CN in the
boundary layer of graphite in nitrogen plasmas. AIAA Paper 2015-2668. 2015.
http://dx.doi.org/10.2514/6.2015-2668.
[47] Park C. Nonequilibrium hypersonic aerothermodynamics. New York, NY: John
Wiley & Sons; 1990.
[48] Zhang L, Pejakovic DA, Marschall J, Dougherty M, Fletcher D. Laboratory
investigation of the active nitridation of graphite by atomic nitrogen.
J Thermophys Heat Tr 2012;26:10e21. http://dx.doi.org/10.2514/1.T3612.
[49] Helber B, Turchi A, Chazot O, Magin T, Hubin A. Gas/surface interaction study
of low-density ablators in sub- and supersonic plasmas. AIAA Paper 2014-
2122. 2014. http://dx.doi.org/10.2514/6.2014-2122.
[50] Scoggins JB, Soucasse L, Riviere P, Soufiani A, Magin T. Coupled flow, radiation,
and ablation simulations of atmospheric entry vehicles using the hybrid sta-
tistical narrow band model. AIAA Paper 2015-3112. 2015. http://dx.doi.org/
10.2514/6.2015-3112.
[51] Munafo
A. Multi-scale models and computational methods for aero-
thermodynamics applications. Ph.D. thesis. Paris, France: Ecole Centrale Paris
and von Karman Institute for Fluid Dynamics; , 2014.
[52] Klomfass A, Müller S. Calculation of stagnation streamline quantities in hy-
personic blunt body flows. Shock Waves 1997;7:13e23. http://dx.doi.org/
10.1007/s001930050057.
[53] Munafo
A, Magin T. Modeling of stagnation-line nonequilibrium flows by
means of quantum based collisional models. Phys Fluids 2014;26. http://
dx.doi.org/10.1063/1.4894842.
[54] Scoggins J, Magin T. Gibbs function continuation for linearly constrained
multiphase equilibria. Combust Flame 2015;162:4514e22. http://dx.doi.org/
10.1016/j.combustflame.2015.08.027.
[55] McBride BJ, Gordon S. Computer program for calculation of complex chemical
equilibrium compositions and applications: I. Analysis. Cleveland, Ohio
44135-3191: National Aeronautics and Space Administration, Lewis Research
Center; 1994. NASA Reference Publication, RP-1311.
[56] Congedo PM, Corre C, Martinez J-M. Shape optimization of an airfoil in a BZT
509
flow with multiple-source uncertainties. Comput Method Appl M 2011;200:
216e32. http://dx.doi.org/10.1016/j.cma.2010.08.006.
[57] Tang K, Congedo PM, Abgrall R. Sensitivity analysis using anchored anova
expansion and high-order moments computation. Int J Numer Meth Eng
2015:1e31. http://dx.doi.org/10.1002/nme.4856.
[58] Archer GEB, Saltelli A, Sobol' IM. Sensitivity measures, anova-like techniques
and the use of bootstrap. J Stat Comput Simul 1997;58:99e120. http://
dx.doi.org/10.1080/00949659708811825.
[59] Yang X, Choi M, Lin G, Karniadakis GE. Adaptive ANOVA decomposition of
stochastic incompressible and compressible flows. J Comput Phys 2012;231:
1587e614. http://dx.doi.org/10.1016/j.jcp.2011.10.028.
[60] X. Wang, On the approximation error in high dimensional model represen-
tation, in: Proceedings of the 2008 winter simulation conference, pp.
453e462, doi: 10.1109/WSC.2008.4736100.
[61] Sobol' I. Theorems and examples on high dimensional model representation.
Reliab Eng Syst Safe 2003;79:187e93. http://dx.doi.org/10.1016/S0951-
8320(02)00229-6.
[62] Xiu D, Karniadakis GE. The Wiener-Askey polynomial chaos for stochastic
differential equations. SIAM J Sci Comput 2002;24:619e44. http://dx.doi.org/
10.1137/S1064827501387826.
[63] Crestaux T, Le Maître O, Martinez J-M. Polynomial chaos expansion for
sensitivity analysis. Reliab Eng Syst Safe 2009;94:1161e72. http://dx.doi.org/
10.1016/j.ress.2008.10.008.
[64] Olynick D, Chen YK, Tauber ME. Aerothermodynamics of the stardust sample
return capsule. J Spacecr Rockets 1999;36. http://dx.doi.org/10.2514/2.3466.
[65] Suzuki T, Fujita K, Ando K, Sakai T. Experimental study of graphite ablation in
nitrogen flow. J Thermophys Heat Tr 2008;22:382e9. http://dx.doi.org/
10.2514/1.35082.
[66] Suzuki T, Fujita K, Sakai T. Experimental study of graphite ablation in nitrogen
flow, part II: further numerical analysis. J Thermophys Heat Tr 2010a;24:
589e97. http://dx.doi.org/10.2514/1.43264.
[67] Suzuki T, Fujita K, Sakai T. Graphite nitridation in lower surface temperature
regime. J Thermophys Heat Tr 2010b;24:212e5. http://dx.doi.org/10.2514/
1.43265.