Minimum Squares PARSEC Parameters for Aerodynamic Airfoils

Ph.D. Carlos Manuel Rodríguez Román (Aeronautical Engineer)

Abstrac
In aerodynamic optimization it is convenient to have a mathematical description of the
geometrical forms that will be tested; the description will be a series of mathematical formulae
which contains several parameters that can be continuously varied in order to find the
extremum of an objective function. The main idea of this work is to find a family of aerodynamic
profiles versatile enough to mimic with sufficient accuracy most members of other families and
to find a procedure to calculate this representation. PARSEC profile characteristics have been
advanced by its creators to promote its use in optimization investigations. In this work, formulas
in matrix form are given to: a) build the profiles of the family, given the parameters; and b) to
adjust the parameters of the family, thru the method of minimum squares, to a given profile.

Introduction

In any aerodynamic optimization procedure it is very convenient to have a mathematical description of the
geometrical forms of the bodies that will be tested. Indeed, the description will be a series of mathematical
formulae which contains several parameters that can be continuously varied in order to find the extremum of an
objective function. In aerodynamics there exist a great number of forms called aerodynamic airfoils that have
been defined in the past by different researchers or different laboratories. Some of these airfoils have been
defined thru mathematical expressions in the form of families [1]. Unfortunately, each family has its own set of
parameters that can be varied to obtain different forms, but there is no relation between one family and the
other.

In an optimization investigation sometimes one finds very convenient to use one type of airfoil in a section of
the body and another type in a different section. If the profiles are not members of the same family, it is very
difficult to find a form to past from one airfoil to the other. To have different sets of parameters could be very
cumbersome and, in the intersections, could arise discontinuities that can be prejudicial for the objective
function (in general, geometrical discontinuities in aerodynamics should be avoided). Thus, it will be very
convenient to have a procedure to represent aerodynamic profiles of different families with a unique set of
parameters.

The main idea of this work is to find a family of profiles versatile enough to mimic with sufficient accuracy most
members of other families. The representation of the profiles by Bezier curves or other general purpose curve
representation techniques could be a good choice, but, as it has been stated by the creators of the PARSEC
profile family [2]: the relation between the parameters and the physical interesting quantities of the profile
(such as lift, drag and moment coefficients) is not clear or direct. These characteristics have been advanced by
its creators to promote the use of the PARSEC profiles in optimization investigations. Then, we will use the
PARSEC family because it works with a set of geometrical parameters that are related to the physical quantities
in a known (or almost known) way and the number of parameters is great enough to give sufficient versatility to
the family. We do not advocate that this is the best choice.

PARSEC Airfoils Definition

The PARSEC profiles are defined by two polynomials and one constraint, the polynomials describes the extrados
and the intrados curves and the constraint is a condition for its contact at the leading edge:

=−

The and are real numbers that have to be found from geometrical conditions imposed to the profile. The
profiles extend from = 0 to = 1 at the leading and trailing edges respectively. Conditions are as
follows for the extrados: position of the maximum, zero slope at the maximum, curvature at the maximum,
position of the trailing edge, and slope at the trailing edge (equivalent conditions are given for the intrados):

= =

1 1
′ =0= − ′ =0= −
2 2
1 3 " 1 3 "
′′ = − − ′′ = − −
2 2 2 2
1 1
+ ∆ = − ∆ =
2 2
1 1 1 1
% & '− ( = − % & '+ ( = −
2 2 2 2

A condition for the radius of the leading edge is added from the radius of curvature of the first polynomial:

1
) =
2

Thus, the set of twelve parameters * , , is replaced by the set of eleven geometrical parameters
*) , , , ′′ , , , ′′ , ,∆ , ', (, and condition = − . Given geometrical
parameters the former equations forms a set of simultaneous linear equations that can be solved for the
unknowns * , ,.
 = -2)
=−
1
3 = 15 35 63 " 99 B 1 ′′ + 8
1 8 0 4 7
4 9:;<
0 4 4 4 4 7 1
0 7
03 5 7 9 11 −
2 " B D
1 8 7 0 2 7
3
0 7 02 2 2 2 2 7 0 7
0 47 = 0 " B D
7
0 57 0 7 0 7
/ 6 0 3 5 7 9 11 7 0% & ' − 1 ( − 1 7
0 2 7 0 2 2 7
2 2 2 2 0 1 7
/ 1 1 1 1 1 6 + ∆ −
/ 2 6
1
3 = 15 35 63 " 99 B
1 ′′ + 8
1 E:EF 8 0 4 7
0
4 4 4 4 4 7 1
2 3 5 7 " 9 B 11 D 0 − 7
1 3
8 0 7 0 2 7
0 7 02 2 2 2 2 7 0 7
0 47 = 0 " B D
7
0 57 0 7 0 7
/ 6 0 3 5 7 9 11 7 0% & ' + 1 ( − 1 7
0 2 7 0 2 2 7
2 2 2 2 0 1 7
/ 1 1 1 1 1 6 − ∆ −
/ 2 6

In an appendix to this work, MATLAB code is given to solve these equations.

Minimum Squares Procedure

The PARSEC family could represent almost any profile of another family if the appropriate geometrical
parameters are given. In order to do so, one has to devise a procedure to calculate these parameters given the
form of the intended profile. When a profile is given, we have or we can calculate a number of points along the
extrados and the intrados of such a profile. The most popular method to adjust a curve to a given set of points is
the method of minimum squares. Thus, given a set of n points by its coordinates GH I , J I K and *H , J , for
L = 1 … N and = 1 … N with & = N + N , the quadratic error commited by representing it trhu the PARSEC
polynomials can be calculated by:

R9 RE
2 2
O= PJ − H Q + PJ − H Q

Now, minimizing this error with respect to parameters * , , will give the closest PARSEC profile to the given
profile. Minimization is achieved by derivating with respect to the parameters and equating to zero:

SO
=0
S* , ,
Introducing the following notation:

R9
1
TU = T
N
RE
1
TUV = T
N

Where T is any quantity evaluated at the extrados and T is any quantity evaluated at the intrados. Equations
for minimization can be written as:

6 6
∂e
(N e X e + N i X i )ae1 + N e
1 1
: aek X ek − N i aik X ik = N e X e2 Ye − N i X i2Yi
∂ae1 k =2 k =2
6 6
∂e 3 ∂e 3
: ae1 X e2 + aek X ek +1 = ye X e2 : ai1 X i2 + aik X ik +1 = Yi X i2
∂ae 2 k =2 ∂ai 2 k =2
6 6
∂e 5 ∂e 5
: ae1 X e3 + aek X ek + 2 = Ye X e2 : ai1 X i3 + aik X ik + 2 = Yi X i2
∂ae 3 k =2 ∂ai 3 k =2
6 6
∂e 7 ∂e 7
: ae1 X e4 + aek X ek + 3 = Ye X e2 : ai1 X i4 + aik X ik + 3 = Yi X i2
∂ae 4 k =2 ∂ai 4 k =2
6 6
∂e 9 ∂e 9
: ae1 X e5 + aek X ek + 4 = Ye X e2 : ai1 X i5 + aik X ik + 4 = Yi X i2
∂ae 5 k =2 ∂ai 5 k =2
6 6
∂e 11 ∂e 11
: ae1 X e6 + aek X ek + 5 = Ye X e2 : ai1 X i6 + aik X ik + 5 = Yi X i 2
∂ae 6 k =2 ∂ai 6 k =2

The last ten equations form two sets of linear algebraic equations that can be written as:

3
X e2 X e3 X e4 X e5 X e6 X e7 ae 2 ye X e2
5
X e3 X e4 X e5 X e6 X e7 X e8 ae 3 Ye X e2
7
ae1 X e4 + X e5 X e6 X e7 X e8 X e9 ae 4 = Ye X e2
9
X e5 X e6 X e7 X e8 X e9 X e10 ae 5 Ye X e2
11
X e6 X e7 X e8 X e9 X e10 X e11 ae 6 Ye X e2
3
X i2 X i3 X i4 X i5 X i6 X i7 ai 2 yi X i2
5
X i3 X i4 X i5 X i6 X i7 X i8 ai 3 Yi X i2
7
ai1 X i4 + X i5 X i6 X i7 X i8 X i9 ai 4 = Yi X i2
9
X i5 X i6 X i7 X i8 X i9 X i10 ai 5 Yi X i2
11
X i6 X i7 X i8 X i9 X i10 X i11 ai 6 Yi X i 2
Introducing the following notation:

3
ae 2 X e2 ye X e2
5
ae 3 X e3 Ye X e2
7
a e = ae 4 ; b e = X e4 ; ce = Ye X e2 ;
9
ae 5 X e5 Ye X e2
11
ae 6 X e6 Ye X e2
X e3 X e4 X e5 X e6 X e7
X e4 X e5 X e6 X e7 X e8
A e = X e5 X e6 X e7 X e8 X e9
X e6 X e7 X e8 X e9 X e10
X e7 X e8 X e9 X e10 X e11

And the corresponding entities for the intrados, equations for minimization can be written in a more compact
form:

(N e xe + N i xi )ae1 + N ebTe a e − N ibTi ai = N e ye xe − N i yi xi

1 1
2 2

ae1b e + A ea e = c e
ai1b i + A iai = ci

And their solution could be easily found to be:

a e = A e−1 (c e − ae1b e )
a i = A i−1 (c i + ae1b i )
1 1
N y x 2 − N i yi xi2 + N i bTi A i−1ci − N e bTe A e−1c e
ae1 = e e e
N e xe + N i xi − N e bTe A e−1b e − N i bTi A i−1b i

Finally, one has to calculate from parameters * , , the geometrical parameters

*) , , , ′′ , , , ′′ , ,∆ , ', (, that correspond to the given profile in order to
make comparisons or to modify parametrically the given profile. In an appendix to this work, MATLAB code is
given to perform these procedures.
APPENDIX - MATLAB Code for PARSEC parameters

Function for finding coordinates of a PARSEC profile from given geometric parameters
function p=parsec(n,pars);

%Dr. Carlos Manuel Rodríguez Román (2009)

ae=zeros(1,6);
ai=zeros(1,6);
ae(1)=sqrt(2*rba);
ai(1)=-ae(1);

b=inv([[3/4*xe^(-1/2),15/4*xe^(1/2),35/4*xe^(3/2),63/4*xe^(5/2),99/4*xe^(7/2)];...
[3/2*xe^(1/2),5/2*xe^(3/2),7/2*xe^(5/2),9/2*xe^(7/2),11/2*xe^(9/2)];...
[xe^(3/2),xe^(5/2),xe^(7/2),xe^(9/2),xe^(11/2)];...
[3/2,5/2,7/2,9/2,11/2];...
[1,1,1,1,1]])*...
[zxxe+1/4*ae(1)*xe^(-3/2);...
-1/2*ae(1)*xe^(-1/2);...
ze-ae(1)*xe^(1/2);...
tan((2*abs-bbs)/2)-1/2*ae(1);...
zbs+1/2*dzbs-ae(1)];
ae(2:6)=b;

b=inv([[3/4*xi^(-1/2),15/4*xi^(1/2),35/4*xi^(3/2),63/4*xi^(5/2),99/4*xi^(7/2)];...
[3/2*xi^(1/2),5/2*xi^(3/2),7/2*xi^(5/2),9/2*xi^(7/2),11/2*xi^(9/2)];...
[xi^(3/2),xi^(5/2),xi^(7/2),xi^(9/2),xi^(11/2)];...
[3/2,5/2,7/2,9/2,11/2];...
[1,1,1,1,1]])*...
[zxxi+1/4*ai(1)*xi^(-3/2);...
-1/2*ai(1)*xi^(-1/2);...
zi-ai(1)*xi^(1/2);...
tan((2*abs+bbs)/2)-1/2*ai(1);...
zbs-1/2*dzbs-ai(1)];
ai(2:6)=b;

t=[0:pi/2/(n-1):pi/2]';
x=1-cos(t);

ze=ae(1)*x.^(1/2)+ae(2)*x.^(3/2)+ae(3)*x.^(5/2)+...
ae(4)*x.^(7/2)+ae(5)*x.^(9/2)+ae(6)*x.^(11/2);
zi=ai(1)*x.^(1/2)+ai(2)*x.^(3/2)+ai(3)*x.^(5/2)+...
ai(4)*x.^(7/2)+ai(5)*x.^(9/2)+ai(6)*x.^(11/2);

p=[flipud([x,ze]);[x,zi]];
end
Function to finding PARSEC geometrical parameters from a given set of points
function p=perf2parsec(perfe,perfi)

%Dr. Carlos Manuel Rodríguez Román (2009)

function y=pparsec(x,a)
y=a(1)*x.^(1/2)+a(2)*x.^(3/2)+a(3)*x.^(5/2)+...
a(4)*x.^(7/2)+a(5)*x.^(9/2)+a(6)*x.^(11/2);
end

function y=dparsec(x,a)
y=1/2*a(1)*x.^(-1/2)+3/2*a(2)*x.^(1/2)+5/2*a(3)*x.^(3/2)+...
7/2*a(4)*x.^(5/2)+9/2*a(5)*x.^(7/2)+11/2*a(6)*x.^(9/2);
end

function y=ddparsec(x,a)
y=-1/4*a(1)*x.^(-3/2)+3/4*a(2)*x.^(-1/2)+15/4*a(3)*x.^(1/2)+...
35/4*a(4)*x.^(3/2)+63/4*a(5)*x.^(5/2)+99/4*a(6)*x.^(7/2);
end

Ae=[[mean(perfe(:,1).^3),mean(perfe(:,1).^4),mean(perfe(:,1).^5),...
mean(perfe(:,1).^6),mean(perfe(:,1).^7)];...
[mean(perfe(:,1).^4),mean(perfe(:,1).^5),mean(perfe(:,1).^6),...
mean(perfe(:,1).^7),mean(perfe(:,1).^8)];...
[mean(perfe(:,1).^5),mean(perfe(:,1).^6),mean(perfe(:,1).^7),...
mean(perfe(:,1).^8),mean(perfe(:,1).^9)];...
[mean(perfe(:,1).^6),mean(perfe(:,1).^7),mean(perfe(:,1).^8),...
mean(perfe(:,1).^9),mean(perfe(:,1).^10)];...
[mean(perfe(:,1).^7),mean(perfe(:,1).^8),mean(perfe(:,1).^9),...
mean(perfe(:,1).^10),mean(perfe(:,1).^11)]];
be=[mean(perfe(:,1).^2);mean(perfe(:,1).^3);mean(perfe(:,1).^4);...
mean(perfe(:,1).^5);mean(perfe(:,1).^6)];
ce=[mean(perfe(:,2).*perfe(:,1).^(3/2));...
mean(perfe(:,2).*perfe(:,1).^(5/2));...
mean(perfe(:,2).*perfe(:,1).^(7/2));...
mean(perfe(:,2).*perfe(:,1).^(9/2));...
mean(perfe(:,2).*perfe(:,1).^(11/2))];

Ai=[[mean(perfi(:,1).^3),mean(perfi(:,1).^4),mean(perfi(:,1).^5),...
mean(perfi(:,1).^6),mean(perfi(:,1).^7)];...
[mean(perfi(:,1).^4),mean(perfi(:,1).^5),mean(perfi(:,1).^6),...
mean(perfi(:,1).^7),mean(perfi(:,1).^8)];...
[mean(perfi(:,1).^5),mean(perfi(:,1).^6),mean(perfi(:,1).^7),...
mean(perfi(:,1).^8),mean(perfi(:,1).^9)];...
[mean(perfi(:,1).^6),mean(perfi(:,1).^7),mean(perfi(:,1).^8),...
mean(perfi(:,1).^9),mean(perfi(:,1).^10)];...
[mean(perfi(:,1).^7),mean(perfi(:,1).^8),mean(perfi(:,1).^9),...
mean(perfi(:,1).^10),mean(perfi(:,1).^11)]];
bi=[mean(perfi(:,1).^2);mean(perfi(:,1).^3);mean(perfi(:,1).^4);...
mean(perfi(:,1).^5);mean(perfi(:,1).^6)];
ci=[mean(perfi(:,2).*perfi(:,1).^(3/2));...
mean(perfi(:,2).*perfi(:,1).^(5/2));...
mean(perfi(:,2).*perfi(:,1).^(7/2));...
mean(perfi(:,2).*perfi(:,1).^(9/2));...
mean(perfi(:,2).*perfi(:,1).^(11/2))];
Ne=length(perfe(:,1));
Ni=length(perfi(:,1));
invAe=inv(Ae);
invAi=inv(Ai);

ae1=(Ne*(mean(perfe(:,2).*perfe(:,1).^(1/2))-be'*invAe*ce)-...
Ni*(mean(perfi(:,2).*perfi(:,1).^(1/2))-bi'*invAi*ci))/...
(Ne*(mean(perfe(:,1))-be'*invAe*be)+Ni*(mean(perfi(:,1))-bi'*invAi*bi));

ae=[ae1;invAe*(ce-ae1*be)];
ai=[-ae1;invAi*(ci+ae1*bi)];

pars(1)=1/2*ae1^2;
pars(2)=1/2*(atan(dparsec(1,ae))+atan(dparsec(1,ai)));
pars(3)=atan(dparsec(1,ai))-atan(dparsec(1,ae));
pars(4)=1/2*(sum(ae)+sum(ai));
pars(5)=sum(ae)-sum(ai);
pars(6)=fzero(@(x) dparsec(x,ae),0.5);
pars(7)=pparsec(pars(6),ae);
pars(8)=ddparsec(pars(6),ae);
pars(9)=fzero(@(x) dparsec(x,ai),0.5);
pars(10)=pparsec(pars(9),ai);
pars(11)=ddparsec(pars(9),ai);

p=pars;

end

References

[1] Abbot, I.H. and von Doenhoff, A.E. (1959) “Theory of Wing Sections” Dover Publications, New York, USA.

[2] Sobieczky H. (1998) “Manual aerodynamic optimization of an oblique flying wing” Proc. Of the 36 Aerospace Sciences
Meeting and Exhibit , Reno, Nevada, USA.