JOURNAL OF AIRCRAFT

Vol. 52, No. 4, July–August 2015

Constrained and Unconstrained Propeller Blade Optimization

Johann Dorfling∗ and Kamran Rokhsaz†
Wichita State University, Wichita, Kansas 67260
DOI: 10.2514/1.C032859
Propeller optimization by means of calculus of variations is addressed in this paper. A procedure for deriving the
Euler–Lagrange equations for both unconstrained and constrained propeller blade-twist optimization is presented.
This same procedure can be followed to optimize other geometric parameters, but blade twist is considered here to
investigate the feasibility of a variable-twist propeller blade. Results from unconstrained twist optimization show that,
regardless of the operating condition (takeoff, climb, or cruise), the maximum efficiency occurs at significantly lower
power coefficients than what a propeller will be required to absorb. Constrained optimization supports this
observation by showing that, for each operating condition, a power constraint lowers the propeller efficiency. Twist
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

distributions obtained from constrained optimization for each of the three operating conditions differ significantly
from one another, but it is shown that the effect on performance is insignificant. This leads to the conclusion that a
variable-twist propeller would not be beneficial for a typical aircraft mission. For the case of a loiter–dash-type
mission profile, however, analysis shows that a variable-twist propeller blade is indeed feasible.

Nomenclature r = radial coordinate from center of propeller

AR = wing aspect ratio T = thrust force
a = speed of sound Tc = thrust coefficient integrand
B = number of blades t = airfoil thickness
b = inverse of ∂Cl ∕∂αi V = relative flow velocity of a blade element
c = airfoil or blade chord V0 = helical speed of a blade element
Cd = D∕1∕2ρV 2 cdr, airfoil drag coefficient v∞ = freestream velocity or aircraft airspeed
CD;0 = aircraft parasite drag coefficient w = induced velocity
CD;i = aircraft induced drag coefficient x = r∕R, nondimensional radial coordinate
CD;tot = aircraft total drag coefficient xhub = nondimensional hub radius
Cd;min = minimum airfoil drag coefficient α = angle of attack
Cdc = compressible airfoil drag coefficient αi = induced angle
Cdi = incompressible airfoil drag coefficient αZL = zero-lift angle of attack
CL = aircraft lift coefficient β = blade angle
Cl = L∕1∕2ρV 2 cdr, airfoil lift coefficient β.75 = blade angle at the 0.75R radial location
Cl;min Cd = airfoil lift coefficient coincident with the minimum γ = ratio of specific heats (equal to 1.4 for air)
drag coefficient η = JCT ∕CP , propeller efficiency
Cl;α = airfoil lift curve slope κ = airfoil technology factor in the Korn equation
Clc = compressible airfoil lift coefficient [Eq. (13)]
Cli = incompressible airfoil lift coefficient λ = Lagrange multiplier
CP = P∕ρn3 D5 , power coefficient ξ = power coefficient integrand multiplier
CPL = power loss coefficient ρ = air density
CT = T∕ρn2 D4 , thrust coefficient σ = Bc∕πr, solidity
D = propeller diameter or drag force τ = thrust coefficient integrand multiplier
e = Oswald efficiency φ = wake helix angle
G0 = Goldstein tip-loss factor φ0 = advance helix angle
J = v∞ ∕nD, advance ratio ω = propeller angular speed, rad∕s
M = V∕a, section relative Mach number
MDD = drag divergence Mach number I. Introduction
n = propeller angular speed, rpm
P
Pc
=
=
power
power coefficient integrand M AXIMIZING the performance of a propeller has been a
concern since the inception of propellers. Drzewiecki [1] in
1900 claimed that placing blade elements at a certain angle of attack
PLc = power loss coefficient integrand
R = propeller-tip radius would result in an optimum propeller. Though his method lacked the
inclusion of induced velocities and did not actually give optimum
propellers, his contribution serves to show the desire of increasing
Presented as Paper 2014-0563 at the 52nd Aerospace Sciences Meeting, the useful output from a propeller. This paper continues the search
National Harbor, MD, 13–17 January 2014; received 12 February 2014; for enhancing the performance of propellers by considering both
revision received 22 April 2014; accepted for publication 10 September 2014; constrained and unconstrained propeller optimization. The focus is on
published online 18 December 2014. Copyright © 2014 by Johann Dorfling propeller blade-twist optimization because the motivation behind this
and Kamran Rokhsaz. Published by the American Institute of Aeronautics and study is to assess the feasibility of a variable-twist propeller from a
Astronautics, Inc., with permission. Copies of this paper may be made for performance aspect. This is purely a performance evaluation, and it is
personal or internal use, on condition that the copier pay the $10.00 per-copy not suggested that a manner for achieving propeller twist change in
fee to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, flight is possible. Thus, this paper introduces a propeller optimization
MA 01923; include the code 1533-3868/14 and $10.00 in correspondence
with the CCC. scheme and then evaluates the optimization results from the standpoint
*Graduate Research Assistant, Department of Aerospace Engineering. of assessing whether or not the variable-twist concept is feasible.
Student Member AIAA. Now, to choose the appropriate optimization method to use in this
† present study, some of the past developments in the field of propeller
Professor, Department of Aerospace Engineering. Associate Fellow
AIAA. optimization will first be reviewed. Perhaps the first formal
1179
 1180 DORFLING AND ROKHSAZ

optimization of a propeller was introduced by Betz [2] in 1919, where approach for the aerodynamics. They separated the optimization
he used a variational approach to arrive at a rigid wake result for problem into two subproblems: one for determining the optimum
minimum induced losses. The ideal Betz wake formed the basis of the twist, and the other for the optimum chord-length distribution. It is
vortex theory of Goldstein [3] as well Theodorsen’s [4] propeller not clear, however, whether they simply optimized the twist and
theory. The design methods outlined by Crigler [5], Larrabee [6], and chord-length sequentially, one after the other, or there was some
Adkins and Liebeck [7] could be considered to be optimization iteration between the two subproblems.
methods because they are used to determine the propeller geometry In 1983, Rizk [16] introduced a numerical optimization approach
that, at a specified operating condition, would have a circulation termed the single-cycle scheme. In a typical optimization procedure,
distribution that gives rise to the ideal Betz wake. An ideal wake the design variables are perturbed toward the objective, and the
implies minimal induced losses and thus maximized efficiency. These aerodynamic tool is rerun to evaluate the new design. Thus, the
approaches do not, however, directly use optimization techniques but aerodynamic analysis is run for each optimization step. What Rizk
instead use the results of Goldstein [3] and Theodorsen [4], who in turn proposed was a single iterative process where, after each perturbation
used the Betz optimized wake [2], to arrive at a design. in the optimization routine, the design variables were updated in the
A more direct optimization approach was introduced by Lock et al. flow solver and a single iteration was performed. The flow solution
[8] in 1942. These authors applied calculus of variations to determine does not converge in this single iteration, but the flow variables are fed
the twist distribution that would minimize the power loss at a constant back to the optimizer, and a next step toward the optimum is made. As
thrust. Moriya [9] also presented a similar variational approach for the design progresses toward the optimum, the number of iterations of
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

determining the optimum twist. Compared to the methods of Crigler the flow solver increases so that, by the end of the optimization process,
[5], Larrabee [6], and Adkins and Liebeck [7], which output both the the flow solver would be at a converged solution using the optimized
twist and chord length distributions, the method of Lock et al. [8] and design variables. This method of Rizk was used along with a potential
Moriya [9] takes the entire blade to be already designed and then flow code of Rizk and Jou [17] and an Euler code developed by Rizk
simply determines the optimum twist at a specified operating condi- [18] and proved to be significantly more efficient.
tion. The advantage of this variational approach is that the twist of an The cases discussed so far all dealt with a single objective,
already designed propeller could be optimized for a different flight typically that of maximizing the efficiency of a propeller, and for the
condition, but this could also be a drawback because the entire most part only considered the aerodynamics in the analysis. But for
propeller geometry needs to already be defined. Calculus of variations many design cases, this may not be sufficient because the propeller
could also be used to determine the optimum chord-length distribution, may have to meet certain noise constraints, and the structural integrity
as was done by Davidson [10] for counter-rotating propellers in 1981. and aeroelastic response of the blade may also need to be taken into
Variational approaches for optimization are very effective but may account. For such multidisciplinary optimizations, the most common
be tedious to apply because they might involve determining either approach appears to be genetic algorithms. This method is initialized
derivatives or integrals in the Euler–Lagrange (E–L) equations. by a population of feasible designs, and through an evolutionary
These methods are also not necessarily always the most efficient process of reproduction, mutation, and crossover, the population
optimization tools. It is thus more common to make use of numerical evolves toward the desired optimum. Lee and Hajela [19] used a
optimization techniques. A large variety of methods have been genetic algorithm for a multidisciplinary design optimization of a
developed and applied to the propeller design problem, with the helicopter rotor blade. The disciplines considered were the
choice of method for a particular optimization hinging on a number of aerodynamics, dynamics, and structure of the rotor blade. In a
factors, including the objective (or objectives), the number of design paper by Burger et al. [20], as well as in Burger [21], a genetic
variables, and the constraints. Shaw [11] developed a program in algorithm was used to optimize the performance of a propeller. The
1970 that used empirical data to optimize a propeller. He used a design variables were the number of blades, the airfoil shape (camber
gradient-based search called a pattern search technique, which of a NACA four-digit airfoil), twist distribution, chord distribution,
searches for the minimum in a direction of descent, but not and blade sweep. The optimization scheme was able to handle both
necessarily in the steepest descent direction, as is often done in aerodynamic and acoustic constraints as well as being capable of
gradient searches. Mendoza [12] also used a gradient search performing two point optimizations (i.e., two advance ratios).
algorithm that was coupled with a propeller blade-element code. The Gur and Rosen [22,23] performed an extensive multidisciplinary
design variables were the blade chord, section thickness, and element design optimization of a conventional propeller wherein they
angle of attack. The objective was to maximize the propeller considered structural, acoustic, and aerodynamic constraints. Their
efficiency without violating a blade bending stress constraint. A aerodynamic model was the simple momentum-blade-element
major disadvantage of the gradient techniques, especially with method. A genetic algorithm was only used as an initial search tool to
nonlinear problems, is the possibility of converging to a local get close to the global minimum. Once the solution was in the
minimum, as opposed to a globally optimized solution. neighborhood of the global minimum, the simplex method was used
Another class of optimization schemes is known as penalty to get closer to the minimum, and then finally the scheme was
function methods, where a constrained optimization problem is switched to the derivative-based steepest-descent gradient search
transformed into an unconstrained problem. The penalty function method to move the result even closer to the optimum. Their method
penalizes any solution that violates a constraint and thus drives the could be applied either to minimize the noise while meeting
solution toward a minimum within the design space. A penalty aerodynamic and structural constraints or to maximize efficiency
function optimization method was used by Chang and Sullivan [13]. subject to noise and structural constraints. Because of their choice of
They used a vortex-lattice (curved lifting-line) code for the propeller aerodynamic model, the method of Gur and Rosen was limited to
aerodynamics and optimized the twist distribution of a propeller to conventional propellers, and their optimization scheme was not
obtain a maximized efficiency while meeting an equality constraint designed to handle multiple design points or multiple objectives.
for the power coefficient. It is not clear from their results what the A recent work by Marinus [24] presented a genetic-algorithm-based
efficiency gain due to the optimization was because they also added multidisciplinary, multi-objective design-optimization scheme. Three
proplets (an addition to propeller blade tips similar to winglets) to the design points were considered, namely takeoff, cruise, and approach,
propeller blades. Chang and Stefko [14] used the exact same and for each point the scheme was used to maximize the efficiency and
approach as Chang and Sullivan [13] for the twist optimization of minimize the sound pressure level, all while meeting structural
high-speed propellers, except they included viscous drag, airfoil constraints. A Reynolds-averaged Navier–Stokes (RANS) solution of
camber, and thickness in the analysis. The new twist obtained differed the propeller flowfield was used for the performance prediction, and
considerably from the baseline design and showed an efficiency the aeroacoustic model used Farassat’s formulation 1A [25], which was
increase of 1%. In these two papers [13,14], the blade-twist derived from the Ffowcs Williams–Hawkings equation. Because a
distribution was considered as the only design variable. Cho and Lee large population made up the design space and RANS simulations are
[15] optimized both the twist and the chord length distributions by time-consuming, an artificial neural network metamodel was trained to
making use of penalty function methods along with a lifting-line predict the performance of the different blade designs that made up the
 DORFLING AND ROKHSAZ 1181

population. To ensure the accuracy of the neural network prediction, it the propeller angular speed ω, so that the power coefficient, defined
was progressively retrained during the optimization process using by Eqs. (4–6), can be obtained:
RANS simulation results of random individuals within the population Z1
[26]. Some interesting designs that showed promising performance CP  Pc dx (4)
characteristics were obtained from this procedure. xhub
Coming now to the end of the literature review, the question of
the appropriate optimization tool for the present study needs to be Pc  ξCl sin ϕ  Cd cos ϕ (5)
addressed. The choice of optimization method hinges on the
objective, the number of design variables, and the constraints. In the
problem being addressed here, the objective is a maximized σπ 2 x 2
ξ J  π 2 x2  (6)
efficiency using a power-equality constraint, and the only design 8
variable is the twist distribution. It is assumed that a propeller will
already have been designed to meet any structural constraints and that The final performance parameter to be defined is the propeller
noise is not an issue and need not be considered as a constraint. efficiency η. This parameter is defined as the ratio of the thrust power,
Accordingly, considering the methods discussed here, the candidate which is the product of the propeller thrust with the aircraft forward
schemes are the calculus of variations approach or the penalty speed Tv∞ to the total power P. In terms of the thrust and power
function method. These two methods are fundamentally the same, coefficients, the efficiency is written as given by Eq. (7):
and either would be appropriate for the present study. Lock et al. [8],
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

CT J
however, made use of the lifting-line/blade-element method for the η (7)
CP
propeller aerodynamics, which is the same aerodynamic prediction
model used here. Thus, calculus of variations is chosen as the The blade-element equations defined previously are not sufficient by
optimization tool for the present study. themselves to give propeller performance parameters because the
wake helix angle depends on the induced angle αi . Therefore, the
II. Method of Analysis blade-element equations need to be coupled with the vortex theory to
determine the induced angle, which, for the purpose of this paper, is
A. Propeller Aerodynamic Model
assumed to be a small angle. The usable form of the vortex theory is
The propeller aerodynamic performance model presented here is the given by Eq. (8), where the factor G0 is the Goldstein tip-loss factor.
blade-element method coupled with the vortex theory of Goldstein [3]. This factor is only found in table form:
Goldstein treated the propeller wake to be a rigid helical wake with a
constant axial displacement velocity along the radial coordinate. In the 8xG0 sin ϕ0 α
Cl  αi  i (8)
blade-element method, a propeller blade is divided into a number of σ b
sections along its span. Each element is then treated as if it is a two-
dimensional airfoil. Figure 1 shows one such blade element along with Equation (8) gives the lift coefficient of a blade-element based on the
the various components of the local flow experienced by it. From Goldstein tip-loss factor, the advance helix angle φ0 , and the induced
Fig. 1, the thrust and torque contribution of each element can be angle. The lift coefficient can also be written in terms of the section
expressed in terms of the elemental lift dL, drag dD, and the wake helix angle of attack, as shown in Eq. (9), where a linear model is assumed:
angle φ. Summing the individual thrust contributions and assuming the
Cl  Cl;α α − αZL  (9)
width of each element to be of differential size, an integral over the
blade span that gives the total thrust of one blade is obtained. The incompressible drag coefficient model is given by Eq. (10) and
Multiplying this integral by the number of blades in the propeller, B, relates the drag coefficient to the lift coefficient:
and writing thrust, lift, and drag in coefficient form, Eqs. (1–3) for the
total propeller thrust coefficient are obtained: Cdi  0.1Cl;min Cd − Cl 4  Cd;min (10)
Z1
CT  T c dx (1) Because propeller-tip speeds are regularly in the transonic flow
xhub regime, it is also necessary to include corrections to the
incompressible lift and drag models that account for the effects of
compressibility. The Kaplan compressibility correction to the lift
T c  τCl cos ϕ − Cd sin ϕ (2) coefficient [27] is used here, defined by Eq. (11):
σπ 2   
τ J  π 2 x2  (3) Clc 1 t∕c 1 1
8  p  p p − 1
Cli 1 − M2 1  t∕c 1 − M2 1 − M2
 2 
1 1
The same is done for the torque to obtain the total propeller torque,  γ  1 −1 (11)
but before casting the integral in coefficient from, it is multiplied by 4 1 − M2

chord line w

V
α

dL αi v
V0
β
ϕ0
ϕ

ω r = 2π nr

dD

Fig. 1 Schematic of the local flow.

 1182 DORFLING AND ROKHSAZ

Drag divergence is modeled by Eq. (12): Taking the derivatives of the modified twist function with respect
( to the perturbation parameter and then setting the perturbation
Cdi  3 M ≤ MDD parameter equal to zero results in Eq. (20):
Cd  (12)
Cdi  1.1 M−M DD
M > MDD R
1−MDD
∂ηe 0 J x1hub CP ∂T c ∕∂β − CT ∂Pc ∕∂βμx dx
  0 (20)
∂ε C2P
In this equation, MDD is the drag divergence Mach number and is
determined from the Korn equation, Eq. (13) [28,29]:
Note that the notation on the left-hand side of Eq. (20) is different from
C t that in Eq. (19) to highlight the perturbation parameter being set equal
MDD  l   κ (13) to zero. This is also the reason for the modified twist function being
10 c replaced by the regular twist function β. Because the auxiliary function
The factor κ is the airfoil technology factor and is equal to 0.87 for μx is arbitrary, Eq. (20) only holds if the term within the parentheses
NACA 6-series airfoils, and κ  0.95 for supercritical airfoils. under the integral sign equals zero. Thus, the E–L equation for the
Before proceeding with the derivation of the Euler–Lagrange (E– unconstrained twist optimization problem is found as Eq. (21):
L) equation for the unconstrained twist optimization, the following
∂T c ∂P
two relations [Eqs. (14) and (15)], which are determined from Fig. 1, CP − CT c  0 (21)
are given because they are needed in the process of solving for the ∂β ∂β
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

optimum twist distribution:

The twist distribution βx that satisfies this equation will give the
ϕ  ϕ0  α i (14) maximum propulsive efficiency at the specified operating condition.
To solve Eq. (21), the derivatives of the thrust and power coefficient
integrands with respect to the twist function β need to be determined.
The differentiation process shown here follows the derivation
β ϕα (15) given by Lock et al. [8]. The equations for the thrust and power
coefficient integrands were given by Eqs. (2) and (5). The derivative
of the thrust coefficient integrand will be considered first. By making
B. Unconstrained Optimization
use of the product and chain rules and treating the drag coefficient Cd
The statement for the unconstrained twist optimization problem as a function of only the lift coefficient Cl , Eq. (22) is obtained:
simply states that the propeller twist distribution βx that maximizes
the efficiency η is to be determined. Efficiency was defined in Eq. (7),  
∂T c ∂ϕ ∂Cl ∂ϕ dCd ∂Cl
and by combining this equation with Eqs. (1) and (4) for the thrust and  τ −Cl sin ϕ cos ϕ−Cd cos ϕ− sin ϕ
∂β ∂β ∂β ∂β dCl ∂β
power coefficient, Eq. (16) is obtained:
(22)
R
J x1 T c x; βx dx
η  R 1 hub (16) This equation can be simplified by introducing the derivatives of
xhub Pc x; βx dx Eqs. (8) and (14), given by Eqs. (23) and (24), respectively:

The Euler–Lagrange equation for this problem will be derived from ∂Cl ∂Cl ∂αi 1 ∂αi
  (23)
Eq. (16). The derivation follows a typical approach as would be found ∂β ∂αi ∂β b ∂β
in most textbooks dealing with calculus of variations such as [30,31].
Note, however, that neither the thrust nor the power coefficient
depends on the derivative of the twist function β. This simplifies the ∂ϕ ∂αi
 (24)
problem considerably. ∂β ∂β
The first step is to introduce a perturbation parameter ε and an
arbitrary auxiliary function μx to obtain a modified twist function Combining Eqs. (23) and (24) and substituting into Eq. (22), the final
βε x, defined by Eq. (17): result for the derivative of the thrust coefficient integrand is obtained
after some simplification as Eq. (25):
βε x  βx  εμx (17)
   
∂T c ∂C ∂C
The auxiliary function is defined to be zero at the end points.  τ l 1 − bCd  cos ϕ − bCl  d sin ϕ (25)
∂β ∂β ∂Cl
Equation (17) is now substituted into Eq. (16) to arrive at Eq. (18):
R Following this exact same procedure for the derivative of the power
J x1hub T c x; βε  dx coefficient integrand, Eq. (26) is obtained:
ηε ε  R 1 (18)
xhub Pc x; βε  dx    
∂Pc ∂C ∂C
 ξ l 1 − bCd  sin ϕ  bCl  d cos ϕ (26)
The extremum of Eq. (18) is obtained by first differentiating it with ∂β ∂β ∂Cl
respect to the perturbation parameter. This step requires the use of
both the quotient and chain rules of differentiation, and the result is Equations (25) and (26) can now be substituted into Eq. (21) to ob-
simplified by making use of Eqs. (1) and (4). This procedure results in tain the Euler–Lagrange equation for the unconstrained twist
Eq. (19): optimization problem [Eq. (27)]:

R1 R1
∂ηε ε JCP xhub ∂T c ∕∂β ε ∂βε ∕∂ε dx − JCT xhub ∂Pc ∕∂βε ∂βε ∕∂ε dx
 0 (19)
∂ε ε0 C2P
 DORFLING AND ROKHSAZ 1183

   
∂C reproduced. The propeller blade airfoil model used in [8] was a Clark
CP τ 1 − bCd  cos ϕ − bCl  d sin ϕ Y airfoil. Sufficient information regarding the airfoil performance
∂Cl
    was found in [8,32–34] so that the results were reproducible. The
∂C only difficulty, and likely source of error, was that many of the airfoil
 CT ξ 1 − bCd  sin ϕ  bCl  d cos ϕ (27)
∂Cl performance parameters were given in graphic form and had to
be converted to tables for use in a computer program. The
In taking the derivatives of the performance coefficient integrands, compressibility corrections used by Lock et al. [8] were also different
the derivative of the drag coefficient with respect to the lift coefficient from those presented here, but these corrections were clearly
appeared. The incompressible drag coefficient Cdi is modeled as presented in [8]. Another important difference was that the power
given by Eq. (10), and the Mach-number dependence of the drag coefficient of Lock et al. needed to be multiplied by 2π to obtain the
coefficient is given by Eq. (12). The derivative of the drag coefficient power coefficient as was defined in this paper.
is thus given by Eq. (28): In [8], the authors solved for the optimum twist of a propeller blade
( dC by minimizing the power loss using a constant-thrust constraint.
dCl  −0.4Cl;min Cd − Cl  M ≤ MDD
di 3
dCd Power loss is easily related to the power and thrust of a propeller.
 dCdi M−MDD 2 1−M (28) Thrust power is defined as the product of the aircraft forward speed
dCl dCl  0.33 1−MDD 4
M > MDD
and the propeller thrust, and the power loss is simply the difference
between the supplied power and thrust power. Equation (33) gives
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

this relationship in coefficient form:

C. Constrained Optimization CPL  CP − JCT (33)

The goal in optimizing the twist of a propeller blade is to maximize
the efficiency, but there are a variety of ways to define the problem to The limits of the integrals that define these three coefficients are the
reach this goal. For example, the efficiency can be maximized at same, and so the integrands are related as given by Eq. (34):
constant power, or the power could be minimized at constant thrust.
Both of these approaches result in the maximization of efficiency. PLc  Pc − JT c (34)
The problem considered here will be to find the twist distribution
such that the thrust coefficient is maximized subject to the constraint Differentiating Eq. (34) with respect to the twist distribution β results
that the power coefficient remains equal to the constant H. This in Eq. (35):
problem is solved by introducing a Lagrange multiplier λ to form a
new functional I, as shown in Eq. (29): ∂PLc ∂Pc ∂T
 −J c (35)
∂β ∂β ∂β
I  CT − λCP − H (29)
The relationship given by Eq. (35) was used to relate the Lagrange
Substituting the integrals that define the thrust and power coefficients multiplier from Lock et al. [8], λL , to the Lagrange multiplier used
[Eqs. (1) and (4)] into Eq. (29) results in Eq. (30): here. Equation (36) gives this relationship between the two Lagrange
Z1 Z1 multipliers, which allowed for the results published by Lock et al. to
I T c − λPc − h dx  Xx; β dx (30) be compared with those obtained from the iteration procedure used
xhub xhub here:

The constant h is defined so that its integral from the hub to the blade 1
tip is equal to the power constraint H. At this point, the problem has λ (36)
J  2πλL
been altered into determining the twist function βx that maximizes
the functional I. The E–L equation for this problem is thus found as For the comparisons presented in Table 1, the Lagrange multiplier
Eq. (31): published by Lock et al. [8] was used to determine the equivalent
Lagrange multiplier from Eq. (36), which was then used to solve for
∂X ∂T ∂P the twist. This twist distribution was then used to determine the
0 c−λ c (31)
∂β ∂β ∂β torque coefficient and efficiency. These two values are compared
with the results from Lock et al. [8] in Table 1. Agreement between
Substituting Eqs. (25) and (26) into Eq. (31) produces the final result the results from the present method and that of Lock et al. was very
for the E–L equation of the constrained twist optimization problem good, with the percent differences in the efficiencies being for the
[Eq. (32)]: most part less than 2%. This difference can be attributed to the
    tabulation of graphical airfoil data and perhaps also to a difference in
∂C rounding while doing calculations by hand. This work [8] was
τ 1 − bCd  cos ϕ − bCl  d sin ϕ
∂Cl published in 1942, and so the calculations were likely carried out by
    hand. The disparities, however, were very small, and it can safely be
∂C
 λξ 1 − bCd  sin ϕ  bCl  d cos ϕ (32) said that the method used for determining the wake helix angle is
∂Cl satisfactory.

B. Optimization Results
Three operating conditions were defined for studying the optimum
III. Results and Discussion twist of a propeller. Once the optimum twist for a particular operating
A. Validation of Results condition was obtained, the performances at the other two design
Equations (27) and (32) give the optimum unconstrained and points were analyzed to compare the three twist distributions.
constrained blade angle distributions though the twist βx is not The advance ratio, propeller angular speed, altitude, and power
explicitly present in either of these two equations. The solutions thus coefficient associated with the three design conditions of takeoff,
require the coupling of these equations with Eqs. (8), (9), (14), and climb, and cruise are given in Table 2. The airspeed associated with
(15). Part of the solution for the twist distribution requires an iterative the takeoff condition was that at takeoff rotation. The values in
procedure to solve for the wake helix angle φ. To validate the Table 2 were determined from [35], in which the operating
procedure used here to iterate for the wake helix angle, results from a environment of the propellers on Beech 1900D airliners were
similar optimization procedure published by Lock et al. [8] was defined. These aircraft were fitted with four-bladed, 110-in.-diam
 1184 DORFLING AND ROKHSAZ

Table 1 Comparison with results from [8]

Lagrange multiplier λL Method CQ η Percent difference η
v∞  352 mph, J  1.95, altitude  21; 000 ft
0.010 Lock et al. [8] 0.0085 0.8390 1.63
0.010 Present 0.0090 0.8253 1.63
0.020 Lock et al. [8] 0.0149 0.8890 0.90
0.020 Present 0.0150 0.8810 0.90
0.039 Lock et al. [8] 0.0252 0.9000 0.14
0.039 Present 0.0254 0.8987 0.14
0.058 Lock et al. [8] 0.0330 0.8980 0.55
0.058 Present 0.0334 0.8930 0.55
0.085 Lock et al. [8] 0.0407 0.8790 0.20
0.085 Present 0.0405 0.8808 0.20
v∞  428 mph, J  2.62, altitude  21; 000 ft
0.049 Lock et al. [8] 0.0353 0.8940 0.29
0.049 Present 0.0356 0.8914 0.29
0.070 Lock et al. [8] 0.0456 0.8900 0.06
0.070 Present 0.0448 0.8905 0.06
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

0.092 Lock et al. [8] 0.0526 0.8830 0.22

0.092 Present 0.0518 0.8849 0.22
0.113 Lock et al. [8] 0.0574 0.8760 0.20
0.113 Present 0.0572 0.8778 0.20
0.133 Lock et al. [8] 0.0616 0.8690 0.28
0.133 Present 0.0612 0.8714 0.28
0.160 Lock et al. [8] 0.0666 0.8590 0.30
0.160 Present 0.0665 0.8616 0.30
v∞  528 mph, J  4.05, altitude  21; 000 ft
0.049 Lock et al. [8] 0.0436 0.8150 3.16
0.049 Present 0.0429 0.7893 3.16
0.097 Lock et al. [8] 0.0605 0.8390 1.55
0.097 Present 0.0600 0.8260 1.55
0.187 Lock et al. [8] 0.0867 0.8370 0.80
0.187 Present 0.0862 0.8303 0.80
0.268 Lock et al. [8] 0.1062 0.8200 0.74
0.268 Present 0.1054 0.8139 0.74
0.340 Lock et al. [8] 0.1222 0.8010 0.45
0.340 Present 0.1217 0.7974 0.45
0.403 Lock et al. [8] 0.1359 0.7850 0.35
0.403 Present 0.1354 0.7822 0.35

conditions. The simplicity of the aerodynamic model does, therefore,

Table 2 Definition of propeller operating conditions not inhibit the usefulness of the results presented here.
Revolutions
Operating condition J per minute Altitude, ft CP;0 1. Unconstrained Problem
Takeoff 0.7173 1725 0 0.1729 The unconstrained problem produces a twist distribution that
Climb 1.2603 1580 7,500 0.2435 maximizes the efficiency of the propeller without regard to the thrust
Cruise 2.0663 1480 15,000 0.3733 and power coefficients at which this maximum efficiency occurs.
Thus, the power coefficient coincident with the maximum efficiency
is not expected to match that given in Table 2. In fact, Table 3
constant-speed propellers. Thus, the propellers optimized here have shows that the power coefficients obtained from the unconstrained
these two parameters in common with the Beech 1900D propellers, optimization were significantly lower at all three advance ratios. This
and it was also assumed that the propellers were of the constant-speed
type, capable of varying their blade angle. Other parameters such as
the chord lengths and thicknesses were estimated; a constant chord 0.45 0.027
length of 9 in. was used for the entire blade span, and the blade was t /c
0.40 0.024
given a thickness that varied from 40% at the root to 5% at the tip.
c/R
Figure 2 shows a plot of the normalized thickness and chord
Thickness and Chord Length

Minimum Drag Coefficient

0.35 0.021
distribution along the blade span. The propeller airfoils were assumed Cd,min
to be symmetric with incompressible lift curve slopes of Cl;α  6.3, 0.30 0.018
and minimum drag coefficients were estimated to vary along the
span, as shown in Fig. 2. These drag coefficient values were assumed 0.25 0.015
to occur at zero angle of attack. As mentioned earlier, compressibility 0.20 0.012
corrections to the incompressible airfoil data were also included.
The propeller aerodynamic model makes use of a simple thin 0.15 0.009
airfoil aerodynamic model and a rigid wake, and it does not include
any installation effects such as angular inflow or flow interaction with 0.10 0.006
a nacelle. For this reason, conclusions should not be drawn from
0.05 0.003
absolute performance results presented here, especially at low-
advance-ratio, high-power operating conditions. However, the 0.00 0.000
objective of this study is not to consider the absolute performance 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
numbers but rather to compare the differences in blade twist Radial Location, x
and performance of propellers optimized for different operating Fig. 2 Propeller blade geometry.
 DORFLING AND ROKHSAZ 1185

Table 3 Unconstrained points will be considered when defining a mission for a variable-twist
optimization results propeller.
Coefficient Takeoff Climb Cruise
2. Constrained Problem
CT 0.0613 0.0747 0.0937
CP 0.0553 0.1092 0.2202 Before proceeding with presenting the results from the
η 0.7953 0.8626 0.8793 unconstrained twist optimization, the Lagrange multiplier will first
be studied to determine whether this value can be interpreted as a
physical quantity. Observing the behavior of the thrust and power
suggests that, whenever a propeller of fixed solidity is required to coefficients while varying the Lagrange multiplier, it was noticed that
either produce more thrust or absorb more power at a specific these quantities both decreased when the Lagrange multiplier was
operating condition, the efficiency deteriorates. increased. This suggested that the Lagrange multiplier had a direct
Figure 3 shows the twist distributions that resulted from the influence on the blade pitch angle. Figure 4 is a plot that shows the
unconstrained optimization. To be able to compare the twists in the resulting blade pitch angles and twists due to different Lagrange
respective blades apart from the blade pitch angles β0.75 , the blade multiplier values. As can be seen here, as well as in Fig. 5, the primary
angles at the tips were all set equal to zero. All three operating effect of an increasing Lagrange multiplier was a decreasing blade
conditions resulted in blades that had between 40 to 45 deg of twist in pitch angle. A much less evident secondary effect was a change in the
amount of twist in the propeller blade and a slight variation in twist
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

them, but the spanwise distribution of the twist varied significantly.

This difference can be explained by considering the spanwise distribution. It is, however, clear from these results that the Lagrange
variation of the advance helix angle φ0 . Generally speaking, the multiplier controls the blade pitch angle.
advance helix angle will be large at the blade root due to the small An alternate interpretation of the Lagrange multiplier is also
rotational velocity component and comparatively large forward possible if the unconstrained problem is compared with the con-
velocity. Thus, for all three cases, large blade angles are seen at the strained problem, specifically comparing Eqs. (21) and (31). Looking
blade root. Moving farther outboard on the blade, the rotational at these two equations side-by-side suggests that the Lagrange
velocity component becomes larger, and the advance helix angle multiplier could be considered as the ratio of the thrust coefficient to
decreases. The rate at which this angle decreases depends on the the power coefficient. But this relationship was found to only be true
advance ratio (i.e., the magnitude of the forward speed relative to the for the case when the power coefficient constraint was set equal to the
rotational component). At low advance ratios, the advance helix power coefficient that resulted from unconstrained optimization. In
angle decreases rapidly near the root, but out toward the tip, the rate of other words, an alternate interpretation of the value of the Lagrange
decline is much reduced. As the advance ratio increases, the variation multiplier is that, if this parameter equals the ratio of the thrust
of this helix angle along the span approaches linear behavior. Figure 3 coefficient to the power coefficient, the absolute maximum propeller
clearly shows that the optimum blade twists followed the spanwise
variation of the advance helix angle. The low-advance-ratio case
showed about a 20 deg decrease in blade angle inboard of the 40% tip 80
radius location and another 20 deg over the outboard portion. λ = 0.4
70
Examining the high-advance-ratio case, it is seen that the blade twist λ = 0.8
was more linear than the low-advance-ratio case, with much less 60 λ = 1.2
change in the blade twist inboard of the 40% tip radius as compared to λ = 1.6
Blade Angle, deg

the low-advance-ratio propeller. The blade twist for the advance ratio 50 λ = 2.0
at which the aircraft climbed lay in between the cruise and takeoff
advance ratio cases, except for inboard of the 0.3R radial position. 40
This condition also resulted in the largest amount of twist over the
entire span of the blade. 30

In closing this subsection, the two most important observations are

20
pointed out. The first is that the optimum blade-twist distribution
varies considerably with advance ratio, though the total amount of 10
twist in a blade remains almost the same. The second is that the power
coefficients associated with the maximum efficiency state are 0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
considerably lower than the power coefficients at which the propeller
Radial Location, x
will be required to operate at the specified design conditions. These
Fig. 4 Plot showing the effect of λ on βx.

45 50
Takeoff
40 45
Climb
35 Cruise 40
0.75R Blade Angle, deg
Blade Twist Angle, deg

30 35

25 30

20 25

15 20

10 15

5 10

0 5

-5 0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0.00 0.25 0.50 0.75 1.00 1.25 1.50 1.75 2.00
Radial Location, x Lagrange Multiplier, λ
Fig. 3 Unconstrained twist distributions. Fig. 5 Plot of β0.75 as a function of λ.
 1186 DORFLING AND ROKHSAZ

45
Takeoff not have a major influence on the performance. Thus, if an aircraft’s
40
Climb mission is a typical one of takeoff, climb, and a long cruise portion,
35
the best compromise would be to twist the propeller for maximum
Cruise
efficiency in cruise as it will not have a considerable effect on takeoff
Blade Twist Angle, deg

30 and climb performance.

25 Considering now the idea of active twist control, the analysis
suggests that the performance benefit for a typical aircraft mission as
20 has been considered here would be almost nonexistent even though
15 the optimum twist at takeoff is considerably different from that in
cruise. Looking back at the unconstrained results, an interesting
10
observation can, however, be made. Compare the thrust coefficients
5 and the efficiencies from the unconstrained optimization results with
0
those given in Fig. 7. The unconstrained efficiency at the takeoff
advance ratio was considerably higher than that shown for the
-5 constrained case, and the thrust and power coefficients were also
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
Radial Location, x
quite a bit lower. As was already mentioned, the maximum efficiency
occurs at lower thrust and power. This suggests that, if an aircraft’s
Fig. 6 Constrained twist distributions.
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

mission consists of extended loitering segments where the air-

craft would be operating at very low-power settings mixed with
maximum-speed dash segments, a variable-twist propeller may be
viable.

C. Analysis of a Loiter–Dash Mission

For this analysis, an aircraft similar in size and performance to the
DAHER-SOCATATBM 850 was considered. Some specifications of
this aircraft and its propeller are given in Table 4 [36]. The propeller
that was evaluated here had the thickness and chord length
distributions used in the preceding analysis (Fig. 2) and had the same
number of blades, but the diameter was set equal to the diameter given
in Table 4. The analysis started by defining the high-speed cruise
operating condition. The true airspeed, angular speed, and power
constraint for this case are shown in Table 5. These values were used
to optimize the propeller for high-speed dash and obtain the
associated thrust for this flight condition. This thrust value was
assumed to be the thrust required to maintain the desired high cruising
Fig. 7 Constrained optimization performance results. speed and was thus considered to be the total aircraft drag at this
condition. With this information, as well as the aircraft weight
(assumed to be the maximum takeoff weight), the induced drag and
efficiency at the specified operating condition is obtained. For any
aircraft parasite drag coefficients could be estimated. Equation (37)
other case, the efficiency will be lower.
was used to estimate the induced drag, which was then used in
Proceeding now with presenting the optimization results, the blade
Eq. (38) to obtain the aircraft parasite drag. The Oswald efficiency e
twist obtained from the constrained optimization will be considered
was assumed equal to 0.8:
first. For each of the design conditions specified in Table 2, the
resulting blade twist was plotted in Fig. 6. Again, the total amount of
C2L
twist in the propeller blade was around 40 to 45 deg. The spanwise CD;i  (37)
distribution of this twist was again different for each operating πeAR
condition but followed the same trend as was shown in Fig. 3.
Outboard of the 40% tip radius location the twist in the climb prop
matched the twist in the cruise prop closely. Inboard of the 0.4R
CD;tot  CD;0  CD;i (38)
coordinate, the climb propeller showed about 17 deg of twist,
whereas the twist in the cruise propeller changed by approximately
13 deg. The takeoff optimized twist was significantly different from
the other two propellers, again showing a much larger change in twist
over the inboard blade section. Table 4 DAHER-SOCATA TBM 850
Because the three optimized twist distributions shown in Fig. 6 for specifications [36]
the three operating conditions were significantly different, it was
expected that the performance of the three blades at each of the design Parameter Value
conditions would be significantly different. Surprisingly, this was not Dimensions
the case. Figure 7 shows the performance results. Each column Length 34.92 ft
represents the condition for which the propeller was optimized, and Height 14.29 ft
Wing area 193.75 ft2
the rows are the condition at which the propeller was being operated. Aspect ratio 8.216
Because the power coefficient was the constraint, it is not shown in Power plant
this table. Notice that the thrust coefficients in each row do not vary Pratt & Whitney Canada PT6A-66D
by much, and similarly, there is not much change in the efficiency in Shaft horsepower 850 hp
each row either. The highest efficiency in each row is coincident with Propeller
the condition for which it was optimized, but the differences are not Number of blades 4
large, with the cruise condition showing the greatest difference of Diameter 91 in.
about 2% between the takeoff and cruise optimized propellers. This is Speed 2000 rpm
Performance
indeed a very interesting observation. It suggests that, when a Maximum takeoff weight 7394 lb
propeller is required to produce a lot of thrust (or absorb a lot of Maximum cruise speed 320 kt
power), the efficiency simply will not be very high, and the twist does
 DORFLING AND ROKHSAZ 1187

Table 5 Definition of the dash and loiter operating conditions

Flight condition v∞ , kt Revolutions per minute J CP CT CD;tot CD;i CD;0
Dash 300.0 2000 2.003 0.2118 0.0928 0.0137 0.0008 0.0130
Loiter 112.2 2000 0.749 0.0460 0.0490 0.0519 0.0389 0.0130

The loitering condition of an aircraft is the speed that gives the constraint and thrust is maximized, but it is not really possible to
maximum endurance. The lift to drag coefficient relationship that generate a good estimate of the power required to maintain the
defines this condition is a combination that maximizes Eq. (39): loitering speed. For this reason, a different optimization was derived
that minimized the power at a constant-thrust constraint. This allowed
 
C3∕2
L
for using the thrust required as a constraint to optimize the twist and
(39) obtain a power estimate for the loiter condition. This is the loiter
CD max power estimate that is shown in Table 5 in coefficient form.
With the operating environment for the high-speed cruise and
This condition can be estimated as given by Eq. (40):
loiter flight conditions defined and an optimized twist distribution for
each case obtained, the performance of each propeller was evaluated.
C2L
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

 3CD;0 (40) As was described in the analysis, optimization of the dash propeller
πeAR used the power coefficient as a constraint to determine the required
thrust for this condition. However, when analyzing the propellers the
The parasite drag coefficient was obtained from the dash analysis, thrust coefficients shown in Table 5 were used as a constraint so that
and so Eq. (40) thus estimated the lift coefficient at the loiter the required power in each case could be compared. Figure 8 shows
condition. Because the aircraft weight was known, the loiter airspeed the results of the performance evaluation.
could be estimated. At this point, the drag force at loiter could also be A propeller optimized for the loitering flight condition and
calculated so that the thrust required to maintain level flight at the operating at this state showed a significantly lower power require-
loitering airspeed was known. The constrained optimization pre- ment to produce the required thrust compared to the dash optimized
sented in this paper was for the case where power is given as a propeller. The loiter propeller had an efficiency during loiter that was
about 4% higher than the dash propeller operating in the loiter
condition. Considering the high-speed dash operating state, it was
observed that the power required to produce the necessary thrust was
lower for the dash optimized propeller than the loiter propeller, and
the efficiency of the dash propeller was about 3.5% higher. The
difference in the performance of the two propellers can be more
appreciated when considering the shaft horsepower required to
produce the necessary thrust for each flight state. The dash propeller
was estimated to need about 10 additional horsepower to produce the
necessary loitering thrust, but in the high-speed cruise case, the dash
propeller would need about 35 hp less than the loiter propeller.
The analysis indicates that there is a potential to improve an
aircraft’s performance by means of a variable-twist propeller when
the aircraft mission is of the loiter–dash type. But it is important to
also consider the optimum twists for the two operating states to know
how large a change in blade twist would be necessary to realize the
performance gains discussed. Figure 9 is a plot of the two resulting
blade twists. The amount of twist in the two blades was again almost
the same, but the distribution of the twist varied significantly. The
maximum difference between the two blades was about 8 deg around
the 50% tip radius location. Such a large change in blade twist is
perhaps not realistically achievable for a propeller in flight. Another
possibility, however, is to adjust the blade camber. This may perhaps
Fig. 8 Performance results: loiter–dash mission. be more realistic because achieving an effective blade angle change
of 8 deg would not require a very large trailing-edge deflection.
Evaluation of an actuation mechanism or a structural analysis of a
45 variable geometry propeller blade is, however, beyond the scope of
Loiter this paper.
40 Dash
35
Blade Twist Angle, deg

30 IV. Conclusions
25 The topic addressed in this paper is that of propeller optimization,
20 and the method chosen by which to perform the efficiency maximiza-
tion study was calculus of variations. The aerodynamic model made
15 use of the blade-element method and a rigid helical wake assumption.
10 A procedure for deriving the Euler–Lagrange equations for both
unconstrained and constrained propeller blade-twist optimization
5
was presented. This same procedure can be followed to arrive at the
0 appropriate Euler–Lagrange equation for optimization of other
-5
propeller geometric variables such as blade chord or airfoil camber.
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 The example of propeller blade twist was given to be able to look into
Radial Location, x the viability of a variable-twist propeller purely from a performance
Fig. 9 Loiter and dash twist distributions. standpoint.
 1188 DORFLING AND ROKHSAZ

The results of the unconstrained twist optimization study showed [14] Chang, L. K., and Stefko, G. L., “Application of an Optimization
that, for each operating condition (takeoff, climb, and cruise), the Method to High Performance Propeller Designs,” 20th AIAA/
maximum efficiency occurred at power coefficients that were SAE/ASME Joint Propulsion Conference, AIAA Paper 1984-1203,
significantly lower than what the propeller would be required to June 1984.
[15] Cho, J., and Lee, S. C., “Propeller Blade Shape Optimization for
absorb at those states. This suggested that it is simply not possible to Efficiency Improvement,” Computers & Fluids, Vol. 27, No. 3, 1998,
attain high efficiencies during high-power events such as takeoff. pp. 407–419.
Results from the constrained optimization supported this claim, doi:10.1016/S0045-7930(97)00035-2
where it was seen that the takeoff efficiency was significantly lower [16] Rizk, M. H., “The Single-Cycle Scheme: A New Approach to
than the climb and cruise efficiencies. Although the twists obtained Numerical Optimization,” AIAA Journal, Vol. 21, No. 12, 1983,
from the constrained optimization for each of the three operating pp. 1640–1647.
conditions were significantly different, it was shown that the effect on doi:10.2514/3.60164
performance was insignificant. This led to the conclusion that a [17] Rizk, M. H., and Jou, W. H., “Propeller Design by Optimization,” AIAA
variable-twist propeller would not offer a noteworthy boost to Journal, Vol. 24, No. 9, 1986, pp. 1554–1556.
doi:10.2514/3.9479
performance and the concept would therefore not be beneficial for a [18] Rizk, M. H., “Aerodynamic Optimization by Simultaneously Updating
typical aircraft mission. Flow Variables and Design Parameters with Application to Advanced
Taking the unconstrained and constrained optimization results Propellers,” NASA CR-182181, 1988.
together, however, a mission profile where such a variable-twist con- [19] Lee, J., and Hajela, P., “Parallel Genetic Algorithm Implementation in
cept may be beneficial was presented. This mission profile consisted of
Downloaded by QUEENS UNIVERSITY BELFAST on December 22, 2015 | http://arc.aiaa.org | DOI: 10.2514/1.C032859

Multidisciplinary Rotor Blade Design,” Journal of Aircraft, Vol. 33,

extended loitering periods where the aircraft would be operating at a No. 5, 1996, pp. 962–969.
very low-power setting to maximize endurance, but mixed in with the doi:10.2514/3.47042
loitering phases were high-speed dash phases where the aircraft would [20] Burger, C., Hartfield, R. J., and Burkalter, J., “Propeller Performance
operate at maximum speed. Such a mission profile was analyzed, and it Optimization Using Vortex Lattice Method and Genetic Algorithm,”
44th AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper 2006-
was shown that a significant performance boost could be realized if a 1067, Jan. 2006.
propeller had a variable-twist capability. It was therefore concluded [21] Burger, C., “Propeller Performance Analysis and Multidisciplinary
that, from a performance standpoint, a variable-twist propeller blade is Optimization Using a Genetic Algorithm,” Ph.D. Dissertation, Dept. of
indeed feasible for the loiter–dash mission. Aerospace Engineering, Auburn Univ., Auburn, AL, 2007.
[22] Gur, O., and Rosen, A., “Multidisciplinary Design Optimization of a
Quiet Propeller,” 14th AIAA/CEAS Aeroacoustics Conference, AIAA
Paper 2008-3073, May 2008.
References [23] Gur, O., and Rosen, A., “Optimization of Propeller Based Propulsion
[1] Drzewiecki, S., “Des Hélices Propulsives,” Bulletin de l’Association System,” Journal of Aircraft, Vol. 46, No. 1, 2009, pp. 95–106.
Technique Maritime, No. 11, 1900, pp. 89–116. doi:10.2514/1.36055
[2] Betz, A., “Schraubenpropeller mit Geringstem Energieverlust,” [24] Marinus, B. G., “Multidisciplinary Optimization of Aircraft Propeller
Nachrichten der Königlichen Gesellschaft der Wissenschaften zu Blades,” Ph.D. Dissertation, Dept. of Mechnical Engineering, Royal
Göttingen, Mathematisch-Physikalische Klasse, Akademie der Military Academy, Brussels, 2011.
Wissenschaften in Göttingen, Göttingen, Germany, 1919, pp. 193–217. [25] Farassat, F., “Derivation of Formulations 1 and 1A of Farassat,” NASA
[3] Goldstein, S., “On the Vortex Theory of Screw Propellers,” Proceeding TM-2007-214853, 2007.
of the Royal Society London, Series A, Vol. 123, No. 792, 1929, pp. 440– [26] Marinus, B. G., Roger, M., and Van Den Braembussche, R.,
465. “Aeroacoustic and Aerodynamic Optimization of Aircraft Propeller
doi:10.1098/rspa.1929.0078 Blades,” 16th AIAA/CEAS Aeroacoustics Conference, AIAA Paper
[4] Theodorsen, T., Theory of Propellers, McGraw–Hill, New York, 1948. 2010-3850, June 2010.
[5] Crigler, J. L., “Application of Theodorsen’s Theory to Propeller [27] Kaplan, C., “Effect of Compressibility at High Subsonic Velocities on the
Design,” NACA Rept. 924, 1949. Lifting Force Acting on an Elliptic Cylinder,” NACA TN-1118, 1946.
[6] Larrabee, E. E., “Practical Design of Minimum Induced Loss [28] Mason, W. H., “Transonic Aerodynamics of Airfoils and Wings,”
Propellers,” Proceedings of the SAE Business Aircraft Meeting and Configuration Aerodynamics, Virginia Polytechnic Inst. and State
Exposition, Society of Automotive Engineers TP 790585, Warrendale, Univ., Blacksburg, VA, 2006, Chap. 7, http://www.dept.aoe.vt.edu/
PA, April 1979. ~mason/Mason_f/ConfigAero.html [retrieved March 2014].
[7] Adkins, C. N., and Liebeck, R. H., “Design of Optimal Propellers,” [29] Boppe, C. W., “CFD Drag Prediction for Aerodynamic Design,”
Journal of Propulsion and Power, Vol. 10, No. 5, 1994, pp. 676–682. AGARD-AR-256, May 1988, pp. 8.1–8.27.
doi:10.2514/3.23779 [30] Schechter, R. S., The Variational Method in Engineering, McGraw–
[8] Lock, C. N. H., Pankhurst, R. C., and Fowler, R. G., “Determination of the Hill, New York, 1967.
Optimum Twist of an Airscrew Blade by the Calculus of Variations,” [31] Komzsik, L., Applied Calculus of Variations for Engineers, CRC Press,
Aeronautical Research Committee, R&M No. 2088, London, Jan. 1942. Boca Raton, FL, 2009.
[9] Moriya, T., “Formulae for Propeller Characteristics Calculation and a [32] Pankhurst, R. C., and Fowler, R. G., “Calculations of the Performance of
Method to Obtain the Best Pitch Distribution,” Selected Scientific and Two Airscrews for a High-Speed Aeroplane,” Aeronautical Research
Technical Papers, Moriya Memorial Committee, Dept. of Aeronautics, Committee, R&M No. 2021, London, April 1941.
Faculty of Engineering, Univ. of Tokyo, Tokyo, 1959, pp. 43–47. [33] Haines, A. B., and Monaghan, B. A., “High-Speed Lift and Drag Data
[10] Davidson, R. E., “Optimization and Performance Calculation of Dual- for Propeller Performance Calculations,” Aeronautical Research
Rotation Propellers,” NASA TP-1948, 1981. Committee, R&M No. 2036, London, Nov. 1945.
[11] Shaw, R. L., “Computerized Aerodynamic Optimization of Aircraft [34] Pankhurst, R. C., and Haines, A. B., “Account of the Derivation of High-
Propellers,” M.S. Thesis, Dept. of Aeronautical Engineering, Naval Speed Lift and Drag Data for Propeller Blade Sections,” Aeronautical
Postgraduate School, Monterey, CA, 1970. Research Committee, R&M No. 2020, London, Aug. 1945.
[12] Mendoza, J. P., “Propeller Design by Numerical Optimization,” [35] Dorfling, J., “Flight Environment of the Propellers on Commuter
Proceedings of the SAE Business Aircraft Meeting, Society of Automo- Aircraft,” M.S. Thesis, Dept. of Aerospace Engineering, Wichita State
tive Engineers TP 770451, Warrendale, PA, May–April 1977. Univ., Wichita, KS, 2008.
[13] Chang, L. K., and Sullivan, J. P., “Optimization of Propeller Blade Twist by [36] “TBM 850 Pilot’s Information Manual,” DAHER-SOCATA, Tarbes,
an Analytical Method,” AIAA Journal, Vol. 22, No. 2, 1984, pp. 252–255. France, 2012.
doi:10.2514/3.48441