AIAA JOURNAL

Vol. 61, No. 9, September 2023

Fast and Robust Computation of Optimal Rotor Designs

Using Blade Element Momentum Theory

Marius L. Ruh∗ and John T. Hwang†

University of California San Diego, La Jolla, California 92093
https://doi.org/10.2514/1.J062611
Recent years have seen a renewed interest in rotary-wing aircraft, due to the emergence of urban air mobility. In the
conceptual design of such aircraft, fast estimation of rotor performance is important. A challenge is that rotor
performance models require a detailed description of the blade geometry, which is not always available at this stage.
We have developed a new, robust method for rapidly computing optimal rotor designs, which is derived from blade
element momentum theory. In this paper, we derive the rotor design method and prove its convergence for all
reasonable model inputs. We verify the design method by comparing the optimal rotor designs it computes to those
found by a nonlinear programming algorithm. Thousands of numerical experiments yield relative errors of less than
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

13 and 5% in terms of the chord and twist profiles, respectively, and less than 0.3% in terms of efficiency, thrust,
torque, and power coefficients. In addition, the design method computes optimal designs 170 times faster compared to
the nonlinear programming algorithm. Based on these findings, we expect this new rotor design method to be useful in
the conceptual design of rotary-wing aircraft.

Nomenclature has motivated significant research in a number of relevant areas,

ax = axial induction factor including design methods for these eVTOL vehicles.
aθ = tangential induction factor In the conceptual design and sizing of UAM aircraft, one of the
c = sectional chord length necessary tools is a fast and robust method for the prediction of rotor
FM = figure of merit aerodynamic performance. The blade element momentum (BEM)
J = advance ratio method is widely used at this stage of the design process due to its
P; CP = power, coefficient computational efficiency and reasonable accuracy for simple operat-
Q; CQ = torque, coefficient ing conditions (i.e., axial flight and hover). For UAM aircraft, these
R = rotor radius operating conditions arise during cruise and vertical flight, depend-
r = radial distance from the rotor hub ing on the aircraft configuration. The lift-plus-cruise configuration
S = area swept by the rotor (see e.g., [3]) has separate rotors for the vertical and forward flight
T; CT = thrust, coefficient segments. In this case, BEM theory provides a good option for
ux = axial induced velocity at the rotor disk modeling the lift rotors in hover and the propellers in cruise as the
uθ = rotational induced velocity at the rotor disk blades are mostly exposed to axial flight. In this paper, we use the
Vx = axial freestream velocity term “rotor” to refer to hover or axial flight conditions, and neglect
Vθ = rotational velocity of the rotor blade flapping and edgewise (inclined) flight. One challenge is that
W = resultant inflow velocity BEM theory requires a detailed description of the rotor geometry,
α = sectional angle of attack specified by a blade twist and chord distribution. Such information is
η = total efficiency typically not available at this design stage.
η1 = efficiency for axial induced losses This paper presents a fast and robust rotor design method based on
η2 = efficiency for tangential induced losses BEM theory, called the BEM-based ideal-loading design (BILD)
θ = sectional blade twist angle method. The goal is to compute the optimal blade twist and chord
σ = sectional blade solidity distributions that minimize aerodynamic losses given a small number
ϕ = sectional inflow angle of design inputs. The optimal rotor geometry is parameterized in
Ω = rotor rotational speed (rad/s) terms of a reference chord length at a specified radial location, an
ω = wake rotational speed (rad/s) airfoil polar, and an operating condition given by advance ratio.
We document several steps that establish this new design method.
First, we formulate a rotor design optimization problem and apply the
calculus of variations to derive a first-order necessary condition for
I. Introduction optimality. Results show that rotors, operating such that this condition

U RBAN air mobility (UAM) aims to enable on-demand and

affordable transportation in metropolitan areas using electric
aircraft with vertical takeoff and landing (eVTOL) capabilities. Mar-
is satisfied, minimize aerodynamic losses. Second, we propose a
method for finding a solution of the first-order condition, given by
expressions for the rotor-induced velocities in terms of an intermediate
ket studies [1,2] show that the UAM industry could be profitable by efficiency term that captures rotational momentum losses. We also
2030. The excitement and envisioned potential surrounding UAM present equations that allow for the backcomputation of the optimal
twist and chord profiles. Third, we prove that our method can always
Presented as Paper 2021-2598 at the AIAA Aviation 2021 Forum, Virtual, find an optimal blade design for reasonable model inputs. Fourth,
August 2–6, 2021; received 9 November 2022; revision received 17 April based on the results of thousands of numerical experiments, we verify
2023; accepted for publication 4 May 2023; published online 29 June 2023. that optimal blade geometries computed by the BILD method agree
Copyright © 2023 by Marius L. Ruh and John T. Hwang. Published by the with those found by a nonlinear programming algorithm.
American Institute of Aeronautics and Astronautics, Inc., with permission. The remainder of this paper proceeds as follows. We first present a
All requests for copying and permission to reprint should be submitted to literature survey in Sec. II, summarizing research on BEM theory and
CCC at www.copyright.com; employ the eISSN 1533-385X to initiate your
request. See also AIAA Rights and Permissions www.aiaa.org/randp. optimal rotor design approaches. Then, in Sec. III, we review classical
*Ph.D. Student, Department of Mechanical and Aerospace Engineering. BEM theory. If the reader is already familiar with BEM methods, we
† recommend moving to Sec. IV, where we derive the BILD method.
Assistant Professor, Department of Mechanical and Aerospace Engi-
neering. Lastly, in Sec. V, we present the results of this research effort.
4096
 RUH AND HWANG 4097

II. Literature Survey [33] to correct small-angle approximations, improve accuracy, and
A. Blade Element Momentum Theory account for viscous terms. To start the design process, an initial value
for the displacement velocity ratio needs to be specified. This is the
BEM methods are widely used in the low-fidelity analysis of rotors
and wind turbines. Madsen et al. [4,5] compared BEM results with ratio between the velocity of the helicoidal vortex sheet and the
numerical actuator disk (AD) simulations and found close correlation freestream velocity. Adkins and Liebeck [33] state that a value of
between the overall performance predictions of rotors using the two zero is acceptable. The optimal blade shape is computed iteratively
methods. Similarly, Gur and Rosen [6] found that in steady, level until the design produces a displacement velocity ratio that matches
flight, higher-fidelity methods such as lifting-line theory or vortex the initial value as well as the desired thrust or power coefficient.
methods do not offer an advantage over BEM methods, which are According to Adkins and Liebeck, the method is expected to con-
computationally more efficient. Despite its computational efficiency verge in less than five iterations. However, convergence issues exist
and reasonable accuracy for axial flow, BEM theory has several for off-design conditions, which can occur near the blade tip due to
limitations. Bontempo and Manna [7–10] extensively investigated large values of the induction factors, which relate the inflow angle and
errors introduced in momentum theory by comparing results of BEM displacement velocity ratio. There exist numerous variations [34–40]
with AD simulations and found two main sources of errors. The first of the MIL design approach, all of which leverage the Betz condition.
error stems from replacing the integral form of the equations with the
differential form. The second error is due to the linearization of the 2. Other Rotor Design Methods
angular blade velocity in the derivation of the angular momentum An alternative approach to optimal rotor design is to numerically
balance. Due to these errors, BEM theory predicts the local induced solve an optimization problem to maximize or minimize an objective
velocities incorrectly in the hub and tip regions. function (e.g., efficiency or torque) with respect to the rotor geometry.
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

Numerous efforts have been made to improve BEM methods. One One optimization strategy is the calculus of variations, which was used
area of improvement has been the modeling of aerodynamic losses as early as the 1940s [41,42] to optimize the blade twist distribution of
near the hub and tip, first addressed by Glauert [11], who introduced existing rotor designs. In a more recent paper (2015), Dorfling and
the original formulation of the Prandtl tip-loss factor. Shen et al. [12] Rokhsaz [43] proposed a method for deriving and solving the Euler–
analyzed various formulations of the Prandtl tip-loss function and Lagrange equations to perform a blade twist optimization. They used a
proposed a new tip-loss model to address inconsistencies in previous combination of blade element and vortex theory as their aerody-
implementations. Branlard and Gaunaa [13] also developed a tip-loss namic model.
correction model, based on vortex theory. Other areas where BEM A more common approach for optimizing rotor blade geometries is
theory has been improved include low-Reynolds-number flow to use nonlinear programming algorithms. Chang and Sullivan [44]
[14–16], modeling large angles of attack and poststall behavior used a penalty method to find the optimal twist distribution for
[17–20], applications for coned rotors [21], as well as slipstream maximizing efficiency of general aviation propellers subject to a
and wake modeling [22–25]. constant power constraint. Results show an improvement of 1–6% in
Lastly, an important consideration in the implementation of BEM ideal efficiency. A similar study was conducted by Li and Stefko [46] to
theory is robustness. Combining blade element and momentum optimize the twist distribution of two NACA propellers operating at
theory leads to a residual that needs to be solved numerically and higher Mach numbers (up to 0.71), using the same penalty method for
robust convergence is important, particularly in multidisciplinary optimization. The authors found no significant improvement in rotor
design optimization (MDO). Sun et al. [26] investigated the conver- efficiency at these higher Mach numbers. Cho and Lee [47] used the
gence of fixed-point iterative schemes to solve the BEM equations extended linear interior penalty function method to optimize the chord
and proposed an improved version. Masters et al. [27] proposed a and twist distribution of multiple SR and NACA rotors. The best
robust numerical scheme based on Monte Carlo simulations and improvement in efficiency was 2.3% for the SR-2 rotor. Lastly, Burger
sequential quadratic optimization. Ning [28,29] showed how to et al. [48] optimized the number of blades, sweep angle, and airfoil
parameterize the residual in terms of a single variable by introducing shape, in addition to the twist and chord distribution for two rotors,
a minor approximation, reducing the problem to a one-dimensional using a genetic algorithm. All prior work mentioned in this paragraph
root finding algorithm with guaranteed convergence. used the vortex lattice method (VLM) for aerodynamic prediction.

B. Optimal Rotor Design

1. Minimum Induced Loss Methods
III. Review of Classical Blade Element Momentum
Minimum induced loss (MIL) methods are low-fidelity, inverse
design approaches that compute the optimal rotor geometry for a Theory
desired thrust or power coefficient, by leveraging the Betz (1919) [30] We present the BEM equations in a form that applies to both hover
condition for MIL. The Betz condition states that a rotor’s maximum (V x
0) and axial flight (V x > 0). Conventional implementations
efficiency is achieved when the circulation along the blade sheds a do not work in hover due to the use of normalized induction factors,
rigid, helicoidal vortex sheet that moves downstream along the rotor which have a singularity for V x
0. For brevity, we do not include
axis at constant velocity. This condition primarily holds for lightly tip loss factors in the presentation of the BEM equations. However,
loaded propellers, but Theodorsen [31] was able to extend it to for all results shown in this paper, Prandtl’s formulation for tip losses
heavily loaded rotors, by considering the contraction of the slip- was used. We note that, for high-loading cases, Goldstein’s [49]
stream. A design method based on the Betz condition was first formulation will produce more accurate results. Figure 1 shows the
introduced by Larrabee [32] and refined by Adkins and Liebeck geometric interpretation of BEM theory.

Far-field Rotor Wake

Vθ = ΩR θ
Xp
dT dL
Q Xp uθ = ω R W α
ux
Vx T ux ux
p0
dQ
1
Vθ − uθ dD
S0 = π R 02 S = π R2 S = πR 2 2
a) Streamtube for momentum theory b) Blade element analysis
Fig. 1 Geometry for blade element momentum theory.
 4098 RUH AND HWANG

A. Momentum Theory for any angle of attack, which is important during optimization. To
Momentum theory predicts the performance of a rotor by analyz- guarantee a smooth transition between the pre- and poststall regimes,
ing the change in momentum of the air in the axial and tangential we apply a cubic smoothing function to the transition region. More
direction. The rotor is modeled as an AD and its effect is confined information on the training data and the smoothing function can be
to a control volume, called streamtube. By considering the increase found in Appendix A.2. Next, the inflow geometry shown in Fig. 1b
in axial momentum only, the equation for sectional thrust can be can be used to write the thrust and torque as
derived as
1
dT
dL cos ϕ − dD sin ϕ
Cx ρW 2 Bcdr (5)
dT
4πρux ux − V x rdr
4πρV 2x 1  ax ax rdr (1) 2

where the axial induced velocity ux is related to the axial induction dQ 1

dL sin ϕ  dD cos ϕ
Cθ ρW 2 Bcdr (6)
factor as ux
V x 1  ax . It is clear that Eq. (1) in terms of ax is r 2
undefined when V x
0, which is not the case when using the non-
normalized form. A similar expression for sectional thrust can be where c is the airfoil chord length and Cx , Cθ , and W are given by
derived by modeling the increase in pressure across the AD in terms
of the tangential induced velocities, using Bernoulli’s principle. The Cx cos ϕ − sin ϕ Cl
expression is shown in Table 1, denoted dT. ~ Next, the equation for
; and
Cθ sin ϕ cos ϕ Cd
sectional torque can be derived by considering the increase in angular
momentum of the air behind the rotor disk. The result is 1 2
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

W
u2x  V θ − uθ (7)
2
dQ
2πρux uθ r2 dr
4πρV x V θ 1  ax aθ r2 dr (2)

where aθ is the tangential induction factor, defined by Glauert [11] as C. Blade Element Momentum Theory
aθ
uθ ∕2V θ . Like Eq. (1), the torque equation in terms of the
induction factors is undefined in hover. The axial and tangential By equating the expressions for thrust [i.e., Eqs. (1) and (5)] and
induced velocities ux and uθ are sources of aerodynamic inefficien- torque [i.e., Eqs. (2) and (6)] derived from momentum and blade
cies, which can be expressed as element theory, we can write the induced velocities as

2σCθ V θ σCx V θ
Vx V θ − uθ ∕2 uθ
and ux
V x 
η1
; and η2
(3) 4 sin ϕ cos ϕ  σCθ 4 sin ϕ cos ϕ  σCθ
ux Vθ
(8)
where η1 and η2 capture the axial and rotational induced losses,
respectively. The factor of one half is due to our definition of uθ , where Cx and Cθ are given by Eq. (7) and σ is the sectional blade
which is adopted from Glauert [11]. More details on the derivation of solidity, defined as σ
Bc∕2πr for a rotor with B blades. By
these expressions and momentum theory in general can be found in a relating the induced velocities given in Eq. (8) to the inflow geometry
conference paper by the authors [50]. shown in Fig. 1b (i.e., tan ϕ
ux ∕V θ − uθ ∕2), a residual can be
derived, given as
B. Blade Element Theory J
In blade element theory, we discretize the rotor blade radially, Rϕ
σCθ  4 sin ϕ cos ϕ  σCx − 4sin2 ϕ (9)
π
treating each blade element as an infinitesimally thin 2-D airfoil, as
shown in Fig. 1b. We assume that there is no interaction between To solve Eq. (9) for ϕ, we need to know the chord and twist
blade elements. For the results shown in this paper, we neglect distribution, along with the airfoil polar. We solve Eq. (9) via a
compressibility effects and model the lift and drag coefficients as bracketed search algorithm to determine ϕ. Then, we use Eq. (8) to
functions of the angle of attack and Reynolds number, such that find the induced velocities, which allows us to use either the blade
element or the momentum equations to compute thrust and torque.
Cl
fα; Re and Cd
gα; Re (4) We summarize the most important equations from BEM theory in
Table 1 and present common rotor performance metrics in Table 2.
We note that at high advance ratios or tip speeds where compressibility
is a factor, this formulation loses accuracy. However, for many eVTOL
configurations, tip Mach numbers rarely exceed 0.3 and we assume IV. BEM-Based Ideal-Loading Design Method
that such effects are negligible. A more accurate surrogate model that In this section, we present a new method for low-fidelity rotor
takes into account the effect of Mach number is an area for future work. design that is based on BEM theory, called the BILD method. This
For our numerical experiments (Sec. V), we use a surrogate modeling method computes the ideal blade chord and twist distributions that
technique called regularized minimal-energy tensor-product splines minimize the total rotor aerodynamic energy losses given a small
(RMTS) [51,52] to establish the relationships in Eq. (4). To generate number of design inputs. The input variables that need to be specified
the training dataset, we use XFOIL [53] to predict lift and drag are the freestream velocity V x, rotor radius, rotational speed, number
coefficients at various combinations of α and Reynolds number and of blades, a reference radius at which we assume a chord length c, and
apply the Viterna method [19] to extrapolate the airfoil polar to the airfoil polar. We present the new theory that underlies the BILD
α
90°. This extrapolation ensures that the model is well-defined method, which solves the following optimization problem.

Table 1 Summary of blade element momentum theory equations

Theory Thrust Torque Note

dT
4πρux ux − V x rdr 1

Axial momentum N/A Assumes that ux
V x  ux0 
2
Angular momentum dT~
2πρ V − 1 u u rdr dQ
2πρux uθ r2 dr ~
dT derived from Bernoulli
θ
2 θ θ
Blade element 1 1 Assumes known c and θ profile
dT
Cx ρW 2 Bcdr dQ
Cθ ρW 2 Bcrdr
2 2
 RUH AND HWANG 4099

Table 2 Summary of rotor performance metrics

Performance variable Metric/coefficient Note

Thrust T n is rotational speed (rev/s); D is rotor diameter

CT

ρn2 D4
Torque Q
CQ
2 5
ρn D
Power P
CP
3 5
ρn D
VxT Ω is rotational speed (rad/s)
Forward flight: η

ΩQ
Efficiency
T T FM replaces η in hover
Hover: FM

P 2πR2 ρ
Loading T C B∫ R0 cr dr
(disk), T (blade) S is rotor disk area; σ r

S σr πR2

Problem 1 (rotor energy loss minimization): Consider a rotor with Lemma 1 (ideal-loading constant): Let ux solve Problem 1. The
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

a desired thrust T  . The rotor energy loss minimization problem is ideal-loading constant, defined as

minimize Eux ∂Er ∕∂ux

ux λ
− (14)

∂T r ∕∂ux
subject to Tux
T ;
(10)
ux rhub 
Urhub ; is constant spanwise, where the partial derivatives are evaluated at ux .
Proof. Problem 1 is a constrained variational problem that can be
ux rtip 
Urtip solved with the method of Lagrange multipliers. A necessary con-
dition for ux to minimize functional E is that there exists a Lagrange
where E and T are functionals given by multiplier λ such that ux satisfies the Euler–Lagrange equation; i.e.,

Eux

rtip
Er r; ux  dr Tux

rtip
T r r; ux  dr (11) ∂Er d ∂Er ∂T r d ∂T r
and − λ −
0 (15)
rhub rhub ∂ux dx ∂ux0 ∂ux dx ∂ux0

representing the total rate of energy loss and total thrust, respectively. Formulations of the Euler–Lagrange equations for constrained varia-
The remainder of this section is organized as follows. First, in tional optimization problems like Problem 1 are treated in various
Sec. IV.A, we use concepts from the calculus of variations to solve textbooks [54–56]. Here, ux0 is the derivative with respect to the radius
Problem 1. The solution to this infinite-dimensional optimization (i.e., ux0
dux ∕dr). Since Er and T r are not dependent on ux0 ,
problem is given by the ideal-loading condition. Second, in Sec. IV.B, Eq. (15) simplifies to
we use the ideal-loading condition to formulate the BILD method by
developing equations to compute the induced velocities ux and uθ ∂Er ∂T
that minimize the overall energy loss. Third, in Sec. IV.C, we show λ r
0 (16)
∂ux ∂ux
how to backcompute the ideal rotor geometry from the induced
velocities.
and we rearrange to obtain Eq. (14). Since ux is parameterized in
terms of r, Eq. (14) has to hold along the entire rotor span, and
A. Derivation of the Ideal-Loading Condition therefore the ideal-loading constant λ is constant spanwise. □
In this section, we derive the ideal-loading condition. We want to Remark 1: Lemma 1 states the first-order necessary condition for
minimize the rate of energy loss E (i.e., power loss) of a rotor while optimality for Problem 1. Results show (see Sec. V) that if Lemma 1
producing some positive thrust. The definition of the power loss is holds, the corresponding ux indeed minimizes the total rate of energy
input power minus output power and can be written as loss. Hestenes [57] derived sufficient conditions for optimization
problems of this nature, which are also known as “isoperimetric”
Er
ΩQr − V x T r (12) problems due to the integral equality constraint.
Glauert [11] first established the result of Lemma 1, and Ruh and
where Er
dE∕dr, Qr
dQ∕dr, and T r
dT∕dr. Substituting Hwang [50] applied it to develop the BILD method. However, both
the expressions for T r and Qr from Eqs. (1) and (2), respectively, we approaches relied on a logic-based thought experiment, and to the
write the power loss as authors’ knowledge, this is the first rigorous solution of this problem
using the calculus of variations. Next, we derive the analytic expres-
Er
2πrρ V θ ux uθ − 2V x u2x  2V 2x ux (13) sion for λ.
Assumption 2: Each blade element operates at its maximum lift-to-
Assumption 1: The rate of energy loss E due to the tangential drag ratio.
induced velocity uθ is negligible. To derive an analytic expression for λ, we need to evaluate the
We see that Er depends on the radius and the induced velocities, right-hand side of Eq. (14). Taking the partial derivative of Er [i.e.,
Eq. (13)] with respect to ux leads to
which, in turn, depend on the radius. As stated in Assumption 1, we
assume that the power loss is primarily due to the axial induced
∂Er ∂u
velocity ux and neglect the effect of rotational losses due to uθ in the
2πrρ V θ uθ  V θ ux θ − 4V x ux  2V 2x (17)
optimization problem. This simplifies the analysis and allows us to ∂ux ∂ux
write the functionals given by Eq. (11) in terms of ux only. Glauert
[11] found that according to momentum theory, losses due to uθ are at To compute ∂uθ ∕∂ux , we divide the expressions for thrust according
most 4% for large torque coefficients, confirming that Assumption 1 to momentum and blade element theory by the corresponding expres-
is reasonable. We now solve Problem 1. sions for torque and equate the fractions. The result is
 4100 RUH AND HWANG

2ux − V x  Cl cos ϕ − Cd sin ϕ This can be rearranged to write uθ in terms of η2 , such that

(18)
uθ Cl sin ϕ  Cd cos ϕ
uθ
2V θ 1 − η2  (24)
From Assumption 2, we can approximate the right-hand side as Next, we recall Eq. (18) but do not approximate the right-hand side as
1∕ tan ϕ, which can be expressed in terms of ux and uθ according 1∕ tan ϕ because we want the induced velocity ux to capture the effect
Fig. 1b. This allows us to write ∂uθ ∕∂ux as of blade profile drag. This results in
∂uθ 22ux − V x  2ux − V x  Cl V θ − uθ ∕2 − Cd ux

(19)

∂ux V θ − uθ uθ Cl ux  Cd V θ − uθ ∕2
(25)

We substitute this expression into Eq. (17) to obtain Substituting Eq. (24) into Eq. (25) and solving for ux leads to
∂Er 22ux − V x  V x − V θ Cd ∕Cl

2πrρ V θ uθ  V θ ux − 4V x ux  2V 2x (20) ux

∂ux V θ − uθ 2
Taking the partial derivative of T r with respect to ux is straightfor- V x − V θ Cd ∕Cl 2 V x Cd
ward, and the result is   η2 V 2θ 1 − η2  (26)
2 V θ Cl
∂T r
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

4πrρ2ux − V x  (21) Remark 3: The inverse of the lift-to-drag ratio appears in Eq. (26).
∂ux In Assumption 2, we stated that each blade element operates at its
maximum lift-to-drag ratio. To investigate the sensitivity of ux , we
By substituting the expressions for ∂Er ∕∂ux and ∂T r ∕∂ux into plot it as a function of the lift-to-drag ratio for different values of V x ,
Eq. (16), we obtain the analytic form for the ideal-loading constant V θ , and η2 . Based on the results (see Fig. A1 in the Appendix),
as increasing Cl ∕Cd beyond 40 does not significantly change ux , and a
value of Cl ∕Cd of about 20 will result in values of ux that are over
V θ uθ V u
λ
 θ x − Vx (22) 90% of the maximum value. Hence, for sufficiently large lift-to-drag
22ux − V x  V θ − uθ ratios, ux is primarily dependent on V x , V θ , and η2 . This is the case for
most airfoils that would be used for eVTOL applications. However,
Remark 2: To obtain the expression for ∂uθ ∕∂ux given by for low-Reynolds-number applications such as micro-UAVs, the
Eq. (19), we make the assumption that viscous effects due to blade maximum lift-to-drag ratio may be low enough such that it signifi-
profile drag are negligible (see Assumption 2). We note that viscous cantly affects ux .
effects are only neglected in this step (i.e., to derive Eq. (19) from Having established expressions for ux and uθ , we can now rewrite
Eq. (18). This allows us to express λ in terms of the induced the expression for λ given by Eq. (22) in terms of η2 by substituting
velocities only, which is convenient. Since aerodynamic losses Eqs. (24) and (26). After some algebra, we obtain a fourth-order
due to uθ are low, the loss of accuracy when evaluating ∂uθ ∕∂ux polynomial of the form
is small, especially given that the energy functional E is only
modeled in terms of ux . This is supported by the results of a large Pη2 
a4 η42  a3 η32  a2 η22  a1 η2  a0
0 (27)
set of verification studies (see Table 3).
Assumption 3: The induced-velocity profiles ux and uθ satisfy the where the polynomial coefficients are in terms of V x , V θ , Cl , Cd and
following conditions at all spanwise locations, such that λ, and are shown in Appendix A.2. The roots of this quartic equation
represent the local annular efficiency η2, which can be determined at
0 ≤ V x < ux and 0 < uθ < V θ each blade section using a root finding algorithm. Even though there
are four mathematical roots that can be real or complex, Eq. (24)
Assumption 3 is reasonable for all design conditions. This is because, requires that η2 lies between 0.5 and 1 since the tangential induced
in order to produce positive thrust, the axial induced velocity ux needs velocity cannot be greater than the rotational velocity of the rotor. In
to be larger than the freestream velocity V x according to momentum practice, only one root satisfies this requirement. Once the distri-
theory [see Eq. (1)]. The tangential induced velocity uθ cannot be bution of η2 is found, the induced velocities are calculated via
greater than the angular speed of the rotor blade V θ . We note that this Eqs. (24) and (26) from which thrust and torque are determined
assumption is trivial in hover. Given Assumption 3, we can easily using momentum theory. Given thrust or torque, the ideal twist and
show that λ > 0 (see Appendix A.1). chord profiles can be backcomputed (see Sec. IV.C). Since the goal
of the BILD method is to compute the optimal blade geometry,
B. Formulation of the BEM-Based Ideal-Loading Design Method a robust method to compute η2 is important to make the algorithm
In this section, we derive equations for the induced velocities ux reliable.
and uθ that minimize the total energy loss per unit time of a rotor. In
combination with the result of the ideal-loading condition stated by Algorithm 1: BILD method
Eq. (22), these equations form the foundation of the BILD method.
Assuming that we know the ideal-loading constant λ, the two 1: Specify advance ratio, airfoil polar, and reference chord length at some
unknowns in Eq. (22) are ux and uθ . In practice, to determine λ, we spanwise location.
apply the BEM equations once at an arbitrary section along the rotor 2: Given the airfoil polar, compute Cl ∕Cd max at specified spanwise
location.
blade. We assume that the chord length is given and determine the
maximum lift-to-drag ratio Cl ∕Cd based on the airfoil polar of our 3: Perform one BEM analysis (see Sec. III.C) to find ux and uθ at specified
spanwise location.
choice. This allows us to compute ux and uθ at that blade section from
which we find λ. Now, we rewrite the analytic form of λ, given by 4: Compute ideal-loading constant λ using Eq. (22).
Eq. (22), in terms of a single variable that connects ux and uθ . This 5: Solve fourth-order polynomial Eq. (27) to find spanwise distribution η2 .
unifying variable is η2 , which we defined earlier in Sec. III.A and state 6: Compute spanwise distribution of ux and uθ from Eqs. (24) and (26); if
again as more than one airfoil is used, estimate Cl ∕Cd max for each airfoil when
computing ux .
V θ − uθ ∕2 7: Compute thrust and torque from Eqs. (1) and (2).
η2
(23) 8: Backcompute ideal rotor geometry (see Sec. IV.C).
Vθ
 RUH AND HWANG 4101

Assumption 4: Let V x ; V θ ; L∕D > 0. rotational speed of V θ
25 m∕s. The plot confirms that a bracketed
Theorem 1 (convergence of BILD method in axial flow): If search algorithm will indeed converge to the analytic roots shown in
Assumption 4 is satisfied and given that a BEM analysis converges, Fig. 2a. By visual inspection, we can conclude that a bracket still
the BILD algorithm, Algorithm 1, is guaranteed to converge. exists within the interval η2
0.5; 1 when V x
0. Therefore, a
Proof. In Algorithm 1, we need to show that step 5 always con- bracketed root finding algorithm will converge under reasonable
verges. We require η2 to lie between 0.5 and 1. Evaluating Eq. (27) at design conditions in hover.
these values leads to
C. Backcomputation of Ideal Rotor Geometry
2V x The last step of the BILD method is the backcomputation of the
P0.5
−V 3θ  V θ V 2θ  V 2x  and (28)
L∕D ideal-rotor geometry given by the blade chord and twist distribution.
As stated in the previous section, knowing the ideal induced veloc-
2V θ 2V θ ities ux [Eq. (26)] and uθ [Eq. (24)] allows us to calculate thrust and
P1
V x V 2θ λ  Vx λ  2V x (29) torque via momentum theory. Now, we want to backcompute the
L∕D L∕D
chord and twist distribution. We recall from Fig. 1b that the twist
If we can show that P0.5 and P1 always have different signs, angle is given as θ
ϕ  α. Since we know the distributions of the
we guarantee a bracket. Given that λ > 0, all terms in Eq. (28) are induced velocities ux and uθ , we can compute the inflow angle ϕ at
positive, and it follows trivially that P0.5 < 0. Likewise, all terms in each radial section as
Eq. (29) are positive, and P1 > 0. Hence, a bracket always exists,
and a bracketed root finding algorithm for η2 will always converge. ux
ϕ
tan−1
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

(30)
Therefore, the BILD method always converges. The first part of the V θ − uθ ∕2
hypothesis of Theorem 1 concerns Assumption 4, which is reason-
able for most design cases. We know that the ideal-loading constant λ, To determine the angle of attack, we make the assumption that α is
the lift-to-drag ratio, and the rotational velocity V θ are always pos- constant for each airfoil along the blade such that it maximizes the
itive for reasonable design condition. Enforcing that the axial free- lift-to-drag ratio. This is a reasonable assumption for our model as it
stream velocity V x is strictly greater than zero is not always aligns well with the ideal-loading analysis. Knowing α and ϕ, we can
reasonable, because in hover this is not the case. Of course, to compute the twist angle θ. The computation of the chord is straight-
approximate the hover condition one could set V x to a small number forward and involves rearranging either Eq. (5) or Eq. (6) to isolate
greater than zero. However, we will consider the case of V x
0 in chord c. For instance, we can compute c as
more detail. The second part of the hypothesis of Theorem 1 is that a
BEM iteration converges. Implementations of BEM methods with 2dT
c
(31)
provable convergence (e.g., by Ning [28]) exist, and we use a similar Cx ρW 2 Bdr
implementation in designing the BILD method. □
Remark 4 (convergence of BILD method in hover): It is easy to see
that Theorem 3 does not hold when V x
0. While the lower bound V. Numerical Results
of the bracket (i.e., P0.5) will always be negative, the upper bound A. Verification of the BEM-Based Ideal-Loading Design Method
is P1
0. This means that the interval η2
0.5; 1 is no longer To verify the BILD method, we use an open-source Python (https://
guaranteed to contain a bracket. In this case, η2
1 is in fact a github.com/LSDOlab/lsdo_rotor) BEM implementation to solve the
solution to Eq. (27), which is shown in Appendix A.2. Since η2 must same optimization problem as the BILD method numerically with the
be less than 1, there has to be another real root between 0.5 and 1. For SNOPT [58] algorithm. We solve an energy minimization problem
polynomials of degree four and lower, it is possible to find an analytic with 30 chord and twist design variables each.
expression for all roots. It turns out that, in this case, there exists only In a first study inspired by UAM design, the thrust constraint for
one other real root, whose expression can be found in Appendix A.2. the optimization is such that the optimized rotor design produces
Due to the complicated nature of this equation, it is difficult to derive enough thrust to support the weight (725 kg) of an Airbus Vahana, an
explicit conditions, for which this root always remains real. electric UAM concept, in hover. The relevant specifications of this
In Fig. 2a, we plot this root as a function of the angular velocity V θ aircraft were provided by two studies, investigating UAM design
for different values of the ideal-loading constant λ. The lift-to-drag [59,60]. Figure 3 shows the result of the BEM optimization (blue) and
ratio is set to 20. We see that as λ approaches zero, η2 approaches the BILD method (red). The figure-of-merit values obtained with the
one. In Fig. 2b, we plot the quartic polynomial Eq. (27) as a function BILD method and through BEM optimization are 0.767 and 0.766,
of the efficiency η2 for the same values of λ, lift-to-drag ratio, and a respectively, resulting in a relative error of 0.05%.

1.0
0.4

0.2
0.9
Component efficiency η 2

0.0
P(Vx = 0)

0.8 −0.2

−0.4
0.7 λ =0 λ = 25 λ =0 λ = 25
λ =5 λ = 30 −0.6 λ =5 λ = 30
λ = 10 λ = 35 λ = 10 λ = 35
−0.8
0.6 λ = 15 λ = 40 λ = 15 λ = 40
λ = 20 λ = 45 λ = 20 λ = 45
−1.0
10 20 30 40 50 0.75 0.80 0.85 0.90 0.95 1.00
Rotational velocity Vθ Component efficiency η 2
a) η2 computed analytically as a function of Vθ b) Equation (27) as a function of η 2

Fig. 2 Visual inspection of the convergence of BILD in hover.

 4102 RUH AND HWANG

0.10 1.0 0.0125

0.05 0.8 0.0100

blade shape (m) BEM optimization 0.6 0.0075

Cd
Cl
0.00
BILD method Cl BEM optimization
0.4 0.0050
−0.05 Cl BILD method
0.2 Cd BEM optimization 0.0025
Cd BILD method
−0.10
0.0 0.0000
0.2 0.4 0.6 0.2 0.4 0.6
radius (m) radius (m)

30 BEM optimization
BILD method 800
twist distribution (deg)

25

power loss (J/s)

600
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

20
400

15
200 BEM optimization
BILD method
10 0
0.2 0.4 0.6 0.2 0.4 0.6
radius (m) radius (m)
Fig. 3 Verification of the BILD method for a UAM design concept.

We see that the backcomputed blade chord and twist distribu-

tions, shown in the left column, match the results from the SNOPT
optimization reasonably well, with relative errors in terms of the
norm of 8.87 and 4.54%, respectively. We also see that the opti-
mized distributions of the sectional lift and drag coefficients are
mostly constant along the span and vary only near the hub and tip.
This supports the assumption made in the BILD method that the
sectional lift and drag coefficients are constant, such that the lift-to-
drag ratio is maximized.
Next, we summarize the results of a large set of verification studies
in which we sample the input design space of the BILD method and
compare key performance metrics (efficiency, thrust, power, torque
coefficient) with those of converged SNOPT BEM optimizations.
The setup of the numerical experiments is as follows. We perform a
full-factorial sweep of advance ratio (0–3), rotor radius (0.5–1.5 m),
altitude (0–3000 m), and reference chord (0.08–0.22 m) at a constant
RPM of 800 and a normalized reference radius of r∕R
0.5. The
design space is discretized with 10 points in each dimension, equal-
ing 10,000 combinations of the four input variables. For each combi-
nation, we evaluate the BILD method and use the computed thrust as Fig. 4 BILD method vs BEM SNOPT optimization time.
an equality constraint in the SNOPT optimization. The goal of these
numerical experiments is to demonstrate the robustness of the BILD the mentioned quantities. In Table 3, we present additional informa-
algorithm, its computational efficiency, and reasonable accuracy with
tion for other key quantities of interest, divided into hover (a) and
respect to high-level performance metrics.
cruise (b). The largest average errors occur for the total rate of energy
In Fig. 4, we show a scatter plot of the BEM optimization time
loss (23.8%) in cruise and the norm of the blade chord distribution in
versus the BILD method. We see that the BILD algorithm is signifi-
hover (13.0%). Collectively, the results presented in this section
cantly faster than the SNOPT algorithm with an average speed-up in
verify that the BILD method is able to predict the optimal blade
optimization time by a factor of about 170 for converged optimiza-
geometry with reasonable accuracy as well as faster and more
tions. While the SNOPT optimization time is on the order of seconds,
robustly, compared to numerical BEM optimization.
which may seem fast, this time can be significant in large-scale MDO
problems that consider rotor analysis and optimization as a subsys-
tem discipline. In addition, the BILD method converged in all cases B. Parameter Sweeps
while the SNOPT optimizations did not converge in about 2% of the 1. Ideal Rotor Shape Sweeps Without Root Chord Correction
cases (orange markers). Convergence failure is of significant concern In Fig. 6, we show 40 ideal rotor designs corresponding to four
in large-scale MDO problems as a single failure of a subsystem model different operating conditions specified by advance ratio J, which
can cause the whole system model to also fail. Lastly, in Fig. 5, we varies from zero to three across the rows. In the first column, we plot
plot predictions of aerodynamic efficiency as well as thrust, power, the ideal chord distribution as a top-down view of the blade. In the
and torque coefficients obtained with the BILD method against the second column, we plot the ideal blade twist distribution from root to
same quantities found by the SNOPT algorithm for cruise (J > 0). tip. The different curves shown in gray represent different designs by
We show the analogous plot for hover in Fig. A3 in the Appendix. The varying the reference chord length cr from 10 cm (black) to 40 cm
trends are very close to linear with average errors of less than 0.3% for (light gray) at a normalized reference radius of r∕R
0.5.
 RUH AND HWANG 4103
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

Fig. 5 Performance metrics obtained with BILD plotted against the same metrics obtained by a nonlinear programming algorithm for cruise.

Table 3 Summary of 10,000 numerical experiments

Quantity BILD method BEM optimization Relative errors (%) [min, mean, max]
a) Hover
Thrust coefficient CT [min, max] [0.044, 0.326] [0.044, 0.326] [2.20E-12, 6.34E-10, 2.33E-08]
Torque coefficient CQ [min, max] [1.00e-3, 0.030] [1.00e-3, 0.030] [2.59E-04, 0.200, 0.666]
Power coefficient CP [min, max] [0.009, 0.187] [0.009, 0.187] [2.59E-04, 0.200, 0.666]
Figure of merit FM [min, max] [0.695, 0.774] [0.696, 0.774] [2.59E-04, 0.201, 0.670]
Rate of energy loss (kW) [min, max] [0.135, 23.831] [0.167, 24.473] [6.537, 13.379, 27.290]
Blade twist distribution (norm) —— —— [2.650, 8.645, 18.874]
Blade chord distribution (norm) —— —— [5.088, 12.980, 22.887]
b) Cruise
Quantity BILD method BEM optimization Relative errors (%) [min, mean, max]
Thrust coefficient CT [min, max] [0.043, 0.717] [0.043, 717] [4.77E-13, 7.69E-10, 3.63E-08]
Torque coefficient CQ [min, max] [0.003, 0.517] [0.003, 0.516] [1.45E-05, 0.115, 0.764]
Power coefficient CP [min, max] [0.019, 3.250] [0.019, 3.245] [1.45E-05, 0.115, 0.764]
Efficiency η [min, max] [0.408, 0.946] [0.406, 0.943] [1.45E-05, 0.116, 0.758]
Rate of energy loss (kW) [min, max] [0.092, 75.613] [0.110, 73.793] [10.934, 23.843, 51.550]
Blade twist distribution (norm) —— —— [0.615, 3.163, 9.364]
Blade chord distribution (norm) —— —— [4.515, 10.990, 39.375]
c) Convergence and run time
Quantity BILD method BEM optimization
Percent converged 100 97.91
Average run time (s) 0.010 1.715
 4104 RUH AND HWANG

cr =0.1 cr =0.17 cr =0.23 cr =0.3 cr =0.37

cr =0.13 cr =0.2 cr =0.27 cr =0.33 cr =0.4

0.2

blade twist (deg)

blade shape (m)

80

J=3
0.0 70

60
−0.2
50
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
0.2 80

blade twist (deg)

blade shape (m)

J=2
0.0 60

−0.2 40
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

0.2

blade twist (deg)

blade shape (m)

60

J=1
0.0
40

−0.2
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
blade twist (deg)
blade shape (m)

0.2
30

J=0
0.0
20
−0.2
10
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
Fig. 6 Ideal rotor blade geometries computed with the BILD method for different advance ratios J and reference chord lengths cr .

We see that the ideal blade shape in hover (i.e., J
0) is noticeably (determined by k) more heavily near the root and allows the chord
different from that in forward flight. The maximum chord length is near profile to take its original form as we move away from the root. Figure 7
the hub and the chord decreases (approximately linearly) toward the shows the effect of this correction method on the chord distribution. We
tip, where the chord is small. We see that the overall chord profile see that all rotor designs have a more reasonable chord length at the
increases with the reference chord cr , which is to be expected. In terms hub. Structural analysis (e.g., see [61–64]) is needed to verify if such a
of the twist distribution, we see that the twist decreases from about design is indeed feasible. It should be noted that this design change
30 deg at the root to between 10 and 20 deg for zero advance ratio. The only affects the chord profile and not the twist profile.
root and tip twist angles increase with the reference chord length cr . To investigate how this modified geometry changes the aerodynam-
As we increase advance ratio J, the ideal rotor shape changes ics of the rotor, we perform BEM simulations of the original and
notably. The location of maximum chord length moves away from the corrected designs for the same operating conditions and determine
hub as J increases and effectively becomes the value of the reference the percent change in key parameters such as thrust, torque, efficiency,
chord cr . Similarly, the twist distribution also changes with J, with and blade solidity. The results are shown in Fig. A4 in the Appendix.
the blade being more twisted. A notable feature of blades designed for Here we plot the quantities of interest against different values of the
forward axial flight is that the chord becomes very small near the hub, reference chord length cr at different operating conditions J, which we
which makes such a design not feasible. This also occurs for designs vary across the rows. The different curves on the subplots represent
obtained with SNOPT BEM optimizations (see Fig. A2 in the different reference radii. We see that the effect of correcting the thin root
Appendix) if no lower bound is placed on the chord length at the root. chord is most noticeable in blade solidity, which is expected. However,
there is no significant change in the performance parameters (less than
2. Ideal Rotor Shape Sweeps with Root Chord Correction 5% for T and Q and less than 1% for efficiency η and figure of merit
FM). This means that, with the chord correction, the BILD method can
In this section, we propose a solution to the problem of the
compute optimal rotor designs for hover and forward flight.
infeasibly small root chord for forward flight. We adjust the chord
distribution that we obtain from the backcomputation step and take a
3. Performance of Ideal Rotors
weighted average of the reference chord cr and the chord distribution
from the backcomputation, such that In Fig. 8, we present the performance of 280 rotors, designed with
the BILD method. Each data point represents a different blade design.
We plot T; Q; η, and FM against varying values of the reference chord
cicor
e−kr cr  1 − e−kr cibc ; i
1; 2 : : : ; n
i i
(32) length cr . The different curves on each subplot represent varying
normalized reference radii. The operating condition is specified by
where ccor is the corrected chord, cr is the reference chord, k is a advance ratio and varied across subplot rows. All rotors were sized
constant, cbc is the original backcomputed value of the chord, and the with a 1.5 m radius and three blades.
superscript i denotes the radial station. We see that Eq. (32) corrects the We see a clear correlation between the performance parameters
chord near the root by weighting some fraction of the reference chord and the reference chord. In general, rotors with higher blade solidity
 RUH AND HWANG 4105

cr =0.1 cr =0.17 cr =0.23 cr =0.3 cr =0.37

cr =0.13 cr =0.2 cr =0.27 cr =0.33 cr =0.4

0.2

blade twist (deg)

blade shape (m)

80

J=3
0.0 70

60
−0.2
50
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)

0.2 80

blade twist (deg)

blade shape (m)

J=2
0.0 60

−0.2 40
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

0.2

blade twist (deg)

blade shape (m)

60

J=1
0.0
40

−0.2
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
blade twist (deg)

0.2
blade shape (m)

30

J=0
0.0
20
−0.2
10
0.2 0.3 0.4 0.5 0.6 0.7 0.2 0.3 0.4 0.5 0.6 0.7
radius (m) radius (m)
Fig. 7 Ideal rotor blade geometries computed with the BILD method for different advance ratios J and reference chord lengths cr , using the root-chord
correction of Eq. (32).

produce more thrust but also require higher torque, which leads to a for high advance ratios will have an efficiency value that is closer to
decrease in efficiency. This trend is confirmed as thrust and torque the maximum achievable value than rotors that are designed for low
increase (mostly linearly) with increasing reference chord lengths for advance ratios. This is most obvious in the design case of hover, in
all advance ratios. For efficiency, the opposite is true, where an which efficiency is zero by definition. However, we can achieve some
increase in reference chord causes a decrease in efficiency and figure positive efficiency if we operated the same design for J > 0.
of merit in the hover case. Another observation is the fluctuating Given that the BILD method computes the optimal blade geometry
values of the quantities of interest. We attribute these fluctuations to for a given design condition, it is important to know how well a design
imperfections in the airfoil surrogate model used in our BEM code performs when changing the operating condition. Figure 9 can be
(see Sec. III.B). Figure 8 serves as a reference to gauge the effects of used to evaluate the sensitivity of an optimal rotor design (in terms of
changing reference chord and radius along with advance ratio on efficiency) to a change in advance ratio. Rotors that are designed for
rotor performance. higher advance ratios are less sensitive to a change in advance ratio.
Lastly, in Fig. 9, we plot efficiency η against advance ratio J for For instance, a rotor designed for Jdes
2 (dark red curve) can
four different ideal blade designs, computed with the BILD method achieve an efficiency of over 80% when operating between J ≈ 1.5
for design operating conditions of J des
0.5; 1; 1.5; 2. We also show and J ≈ 3.1. A rotor designed for Jdes
0.5 (blue curve) achieves an
the corresponding blade chord and twist profiles, which we used as efficiency of 80% or higher on a smaller range, between J ≈ 0.7 and
inputs to BEM analyses to produce the η versus J curves. J ≈ 1.1. Knowing how sensitive a rotor’s performance is to a change
The goal is to investigate the sensitivity of blade designs produced in operating conditions is important in vehicle design.
with the BILD method to changes in operating conditions. The circle-
shaped markers show the efficiency of the rotors at the design
operating condition while the star-shaped markers show the maxi- VI. Conclusions
mum efficiency that is attainable with the given rotor design. We see Motivated by a renewed interest in rotary-wing aircraft due to the
that in all cases the maximum efficiency (over 80% for all rotors) is emergence of electrified propulsion, this paper presented a fast and
higher than the efficiency at the design operating condition. This is robust design method for aerodynamically optimal rotors, called the
because the BILD method computes the optimal blade design for a BILD method. The method is based on BEM theory and computes
given design condition, specified by advance ratio. However, this the optimal blade chord and twist distributions that minimize the
does not mean that the highest efficiency will be achieved at that aerodynamic losses for a given operating condition. To verify the
design condition. It is well known that, for a given blade design, BILD method, we conducted 10,000 numerical experiments in which
efficiency increases with advance ratio before reaching a maximum we compared the results of the BILD method with those found by a
value, after which it decreases sharply. This trend can be observed in state-of-the-art nonlinear programming algorithm. Based on the
Fig. 9. In terms of the BILD method, this means that rotors designed results, the BILD method can find the optimal rotor design 170 times
 4106 RUH AND HWANG

r/R = 0.25 r/R = 0.3 r/R = 0.4 r/R = 0.5 r/R = 0.6 r/R = 0.7 r/R = 0.8

Torque Q (N-m)
3000
Thrust T (N)

Efficiency η
4000 0.8

J=3
2000
2000 0.6
1000
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
reference chord length (m) reference chord length (m) reference chord length (m)

3000 0.9

Torque Q (N-m)
Thrust T (N)

Efficiency η
2000 0.8

J=2
2000
0.7
1000
1000 0.6

0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

reference chord length (m) reference chord length (m) reference chord length (m)

2000
Torque Q (N-m)

750
Thrust T (N)

0.8

Efficiency η
1500

J=1
500
0.7
1000
250
500 0.6
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
reference chord length (m) reference chord length (m) reference chord length (m)

0.85

Figure of merit FM
2000 400
Torque Q (N-m)
Thrust T (N)

1500 0.80

J=0
200
1000
0.75
500
0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4 0.1 0.2 0.3 0.4
reference chord length (m) reference chord length (m) reference chord length (m)
Fig. 8 Rotor performance of 280 rotors designed using the BILD method.

Fig. 9 Efficiency η vs advance ratio J for different ideal rotor geometries.

faster with relative errors of less than 0.3% for key aerodynamic Despite these promising results, the BILD method has two main
performance metrics such as efficiency, thrust, torque, and power limitations. First, it only considers the aerodynamics of the rotor and
coefficient. In addition, the BILD method is provably convergent, does not take into account blade flapping, structural dynamics, or
and indeed converged in all cases considered, whereas the nonlinear acoustics, which are important disciplines in comprehensive rotor
programming algorithm failed to converge in 2% of the cases. blade design. Second, the aerodynamic analysis is limited by the
 RUH AND HWANG 4107

accuracy and applicability of BEM theory. As such, the wake is not Cl0  Clα α α ≤ αstall − ϵ
modeled and the effect of edgewise flow cannot be captured with
Cl α
c3 α  c2 α  c1 α  c0 ; αstall − ϵ ≤ α ≤ αstall  ϵ
3 2
reasonable accuracy. Re
const:
The results of this paper suggest that, for appropriate design cos2 α
A1 sin2α  A2 ; α ≥ αstall  ϵ
scenarios, the BILD method is a viable option for fast first-order sinα
approximations of aerodynamically optimal rotor geometries. Such
design scenarios may arise at the conceptual level for axial flow and (A1a)
in cases where blade dynamics are not modeled and acoustic con- Cd α
straints are not enforced. Examples of appropriate design scenarios Re
const:
are certain UAM concepts during cruise (e.g., Lift+Cruise) or mili- a4 α4  a3 α3  a2 α2  a1 α  a0 ; α ≤ αstall − ϵ
tary unmanned aerial vehicles (UAVs). In addition, due to its proven
robustness and fast computation time, BILD may be embedded into
d3 α3  d2 α2  d1 α  d0 ; αstall − ϵ ≤ α ≤ αstall  ϵ
large-scale MDO studies of rotary-wing aircraft. An MDO tool that
has received attention in the last couple of years is NASA’s Open- B1 sin2 α  B2 cos α; α ≥ αstall  ϵ
MDAO framework [65], which uses gradient-based optimization
(A1b)
algorithms to efficiently solve large-scale coupled systems. Two
MDO investigations [66,67] of electric aircraft using OpenMDAO where the smoothing region is confined to ϵ
 1.5° and the Viterna
have been done in which the rotor aerodynamics was modeled using coefficients A1 ; A2 ; B1 ; and B2 can be found in [19]. Smoothing is
BEM theory and the blade geometry was optimized numerically. necessary to avoid training the surrogate model on noisy data and to
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

Performing blade shape optimization with the BILD method can avoid local optima during optimization.
potentially speed up the system-level optimization. Equation (A2) repeats Eq. (27) presented in Sec. IV.B, including
In terms of future work, a comparison of BILD with the well- the polynomial coefficients. The coefficients are expressed in terms
established MIL design method can provide insights into the strengths of advance ratio J (J
V x π∕V θ ), lift-to-drag ratio (L∕D), and the
and limitations of both methods in terms of the underlying theory, ideal-loading constant λ.
computational efficiency, and robustness. Other areas of future work
include the use of Goldstein’s tip-loss formulation instead of the Pη2 
a4 η42  a3 η32  a2 η22  a1 η2  a0
0;
Prandtl correction for highly-loaded rotors, the integration of BILD
into large-scale MDO studies, exploring other application domains where
such as wind energy and hydrodynamics, and extending the BILD 8π 6 λ2 J 2 π 4 6λJ3 π 3
algorithm to allow for multiple airfoils along the rotor span. For the a0
−4π 6   4J2 π 4  8J4 π 2  
L∕D 2 V 2θ Vθ
results shown in this paper, a single airfoil was used. While the
underlying theory of BILD allows for multiple airfoils, verification 4λπ 6 20Jπ 5 20J 3 π 3 8J2 π 4 14λJ 2 π 4
− − −  −
is needed to ensure that the designs remain optimal. L∕DV θ L∕D L∕D L∕D2 L∕DV θ
2λ2 Jπ 5 4λJπ 5
− 
Appendix: Supplementary Equations and Figures L∕DV θ L∕D2 V θ
2

A.1. Positivity of the Ideal-Loading Constant λ 24π 6 4λ2 π 6 4λ2 J 2 π 4

a1
40π 6 −   16J2 π 4 − 24J 4 π 2 −
We briefly show that, given Assumption 3, the ideal-loading L∕D 2 Vθ2 V 2θ
constant is strictly positive. The expression for λ is 24λJπ 5 20λJ3 π 3 12λπ 6 84Jπ 5 84J3 π 3
 −   
Vθ Vθ L∕DV θ L∕D L∕D
V θ uθ V u
λ
 θ x − Vx 24J 2 π 4 68λJ 2 π 4 12λ2 Jπ 5 16λJπ 5
22ux − V x  V θ − uθ −   −
L∕D2 L∕DV θ L∕DV 2θ L∕D2 V θ
The first term is greater than zero; i.e.,
16π 6 20λ2 π 6 4λ2 J 2 π 4
a2
−132π 6  − − 116J 2 π 4  16J4 π 2 
V θ uθ L∕D 2 2
Vθ V 2θ
0<
22ux − V x  104λJπ 5 16λJ3 π 3 8λπ 6 128Jπ 5 128J3 π 3
−  − − −
This implies that λ is greater than the last two terms, such that Vθ Vθ L∕DV θ L∕D L∕D
16J2 π 4 112λJ2 π 4 24λ2 Jπ 5 16λJπ 5
V θ ux  − − 
λ> − Vx L∕D2 L∕DV θ L∕DV 5θ L∕D2 V θ
V θ − uθ
32λ2 π 6 144λJπ 5 64Jπ 5 64J3 π 3
a3
160π 6   160J2 π 4   
Since the term V θ ∕V θ − uθ  > 1, we can remove it and maintain the 2
Vθ Vθ L∕D L∕D
inequality, which leads to
64λJ2 π 4 16λ2 Jπ 5
 
λ > ux − V x > 0 L∕DV θ L∕DV 2θ
16λ2 π 6 64λJπ 5
a4
−64π 6 − 64J2 π 4 − − (A2)
A.2. Supplemental Equations V 2θ Vθ
Equation (A1) shows the piecewise formulation of the lift and drag
Equation (A3) is a special case of the quartic Eq. (A2), when V x
0.
coefficient for a single-Reynolds-number case. This formulation is
It is clear that η2
1 is a solution.
repeated based on XFOIL data for a total of ten Reynolds numbers
ranging from 5 × 104 to 2 × 106 . The raw XFOIL data are interpo-
lated such that, within the stall regime, the lift coefficient is modeled PV x
0
−4V θ η2 − 1 4λ2  16V 2θ η32 − 4λ2  24V 2θ η22
linearly with α and the drag coefficient is interpolated with a fourth-
order polynomial. Cubic smoothing is applied in the transition 4V 2θ 2V λ 2V 2θ V λ
 λ2  9V 2θ −  θ η2 − V 2θ  − θ (A3)
region, and the Viterna [19] method is applied in the poststall region L∕D 2 L∕D L∕D2 L∕D
to extrapolate to 90°. Once the data are properly interpolated for
each Reynolds number case, two surrogate models are trained for Cl Equation (A4) shows the expression for the second real root of
and Cd as a function of Reynolds number and angle of attack. Eq. (A2). Recall that the first real root is η2
1.
 4108 RUH AND HWANG

d A.3. Supporting Figures

η2
A  B  ;
B Figure A1 shows the sensitivity of the normalized axial induced
where velocity ux [see Eq. (26)] as a function of the lift-to-drag ratio for
λ2  6V 2θ 1∕3 different values of V x , V θ , and η2 .
A
B
a a  b − c2 − d3  b − c Figure A2 shows additional blade profiles similar to Fig. 3,
3λ2  12V 2θ
obtained with the BILD method and the SNOPT BEM optimization,
where for advance ratios of J > 0. We see that the small chord length at the
0.1250V θ V θ L∕D2  λL∕D − 2V θ  root also occurs for the SNOPT BEM optimization for larger values
a
of advance ratios.
L∕D2 λ2  4V 2θ 
Figure A3 shows additional verification results of the BILD
0.0370λ2  6V 2θ 3 method by plotting key performance metrics (efficiency, thrust,
b

λ2  4V 2θ 3 power, torque coefficient) against the same values obtained by a
0.0417λ2  6V 2θ L∕D2 λ2  9L∕D2 V 2θ  2L∕DλV θ − 4V 2θ  nonlinear programming algorithm. These results are a subset of the
c
10,000 numerical experiments in which we perform a full-factorial
L∕D2 λ2  4V 2θ 2
sweep of key input parameters as explained in Sec. V.A. The operat-
L∕D2 λ4  9L∕D2 λ2 V 2θ  36L∕D2 V 4θ ing condition considered is hover.
0.0278 Figure A4 shows the effect of the root-chord correction technique
−6L∕Dλ3 V θ − 24L∕DλV 3θ  12λ2 V 2θ  48V 4θ
d
(A4) proposed in Sec. V.B.2. It is clear that the change in thrust, torque, and
L∕D2 λ2  4V 2θ 2
efficiency is negligibly small. The biggest effect is the increase in
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

Fig. A1 Normalized ux vs lift-to-drag ratio for different values of V x , V θ , and η2 .

J=0.5

J=1.0

J=2.0

J=3.0

Fig. A2 Blade geometries obtained with BEM optimization and BILD for J > 0.
 RUH AND HWANG 4109
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

Fig. A3 Performance metrics obtained with BILD plotted against the same metrics obtained by a nonlinear programming algorithm for hover.

r/R = 0.25 r/R = 0.3 r/R = 0.4 r/R = 0.5 r/R = 0.6 r/R = 0.7 r/R = 0.8

0.5
10
2
∆ Q (%)

∆ σ (%)
∆ η (%)

0.0
∆ T (%)

J=3
−0.5 5
0 0
−1.0
0
0.2 0.4 0.2 0.4 0.2 0.4 0.2 0.4
reference chord length (m) reference chord length (m) reference chord length (m) reference chord length (m)

10
2
2 0.0
∆ Q (%)

∆ σ (%)
∆ η (%)
∆ T (%)

J=2

1
5
0 0
−0.5
−1
0.2 0.4 0.2 0.4 0.2 0.4 0.2 0.4
reference chord length (m) reference chord length (m) reference chord length (m) reference chord length (m)

0.2 7.5
2
2
0.0 5.0
∆ Q (%)

∆ σ (%)
∆ η (%)
∆ T (%)

1
J=1

2.5
0 0 −0.2
0.0
−1 −0.4
0.2 0.4 0.2 0.4 0.2 0.4 0.2 0.4
reference chord length (m) reference chord length (m) reference chord length (m) reference chord length (m)

4
2
∆ FM (%)

0.0
∆ Q (%)

∆ σ (%)
∆ T (%)

2 0
J=0

0 0
−0.2 −5
−2
0.2 0.4 0.2 0.4 0.2 0.4 0.2 0.4
reference chord length (m) reference chord length (m) reference chord length (m) reference chord length (m)
Fig. A4 Percent change in T, Q, η, and σ due to root-chord correction.
 4110 RUH AND HWANG

blade solidity, which is expected. We believe that the fluctuations in Jan. 2011.
some of the subplots are due to imperfections in the airfoil surrogate https://doi.org/10.2514/6.2011-1255
model (see Sec. III.B). [17] Tangler, J., and Kocurek, D., “Wind Turbine Post-Stall Airfoil Perfor-
mance Characteristics Guidelines for Blade-Element Momentum Meth-
ods,” 43rd AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper
Acknowledgment 2005-0591, Jan. 2005.
https://doi.org/10.2514/6.2005-591
This work was supported, in part, by NASA Grant [18] Whitmore, S. A., and Merrill, R. S., “Nonlinear Large Angle Solutions
No. 80NSSC21M0070. of the Blade Element Momentum Theory Propeller Equations,” Journal
of Aircraft, Vol. 49, No. 4, July 2012, pp. 1126–1134.
https://doi.org/10.2514/1.c031645
References [19] Mahmuddin, F., Klara, S., Sitepu, H., and Hariyanto, S., “Airfoil Lift and
[1] Hasan, S., “Urban Air Mobility (UAM) Market Study,” HQ-E-DAA- Drag Extrapolation with Viterna and Montgomerie Methods,” Energy
TN70296, NASA, Washington, D.C., 2019, https://ntrs.nasa.gov/ Procedia, Vol. 105, May 2017, pp. 811–816.
citations/20190026762. https://doi.org/10.1016/j.egypro.2017.03.394
[2] Goyal, R., Reiche, C., Fernando, C., Serrao, J., Kimmel, S., Cohen, A., [20] Montgomerie, B., Methods for Root Effects, Tip Effects and Extending
and Shaheen, S., “Urban Air Mobility (UAM) Market Study,” HQ-E- the Angle of Attack Range to {+-} 180 deg., with Application to Aero-
DAA-TN65181, NASA, Washington, D.C., 2018, https://ntrs.nasa.gov/ dynamics for Blades on Wind Turbines and Propellers, Swedish
citations/20190000519. Defence Research Agency, Stockholm, Sweden, 2004.
[3] Silva, C., Johnson, W. R., Solis, E., Patterson, M. D., and Antcliff, K. R., [21] Crawford, C., “Re-Examining the Precepts of the Blade Element
“VTOL Urban Air Mobility Concept Vehicles for Technology Develop- Momentum Theory for Coning Rotors,” Wind Energy, Vol. 9, No. 5,
2006, pp. 457–478.
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

ment,” 2018 Aviation Technology, Integration, and Operations

Conference, AIAA Paper 2018-3847, June 2018. https://doi.org/10.1002/we.197
https://doi.org/10.2514/6.2018-3847 [22] Rwigema, M. K., “Propeller Blade Element Momentum Theory with
[4] Madsen, H. A., Bak, C., Døssing, M., Mikkelsen, R., and Øye, S., Vortex Wake Deflection,” 27th International Congress of the Aeronaut-
“Validation and Modification of the Blade Element Momentum Theory ical Sciences, Vol. 2010, International Council of the Aeronautical
Based on Comparisons with Actuator Disc Simulations,” Wind Energy, Sciences, Nice, France, 2010, p. 23.
Vol. 13, No. 4, May 2010, pp. 373–389. [23] Veldhuis, L. L., “Review of Propeller-Wing Aerodynamic Interference,”
https://doi.org/10.1002/we.359 24th International Congress of the Aeronautical Sciences, Vol. 6, No. 1,
[5] Madsen, H. A., Mikkelsen, R., Øye, S., Bak, C., and Johansen, J., “A ICAS Paper 2004-6.3.1, Yokohama, Japan, 2004.
Detailed Investigation of the Blade Element Momentum (BEM) Model [24] Conway, J. T., “Analytical Solutions for the Actuator Disk with Variable
Based on Analytical and Numerical Results and Proposal for Modifi- Radial Distribution of Load,” Journal of Fluid Mechanics, Vol. 297,
cations of the BEM Model,” Journal of Physics: Conference Series, Aug. 1995, pp. 327–355.
Vol. 75, July 2007, Paper 012016. https://doi.org/10.1017/s0022112095003120
https://doi.org/10.1088/1742-6596/75/1/012016 [25] Ning, A., Hayman, G., Damiani, R., and Jonkman, J. M., “Development
[6] Gur, O., and Rosen, A., “Comparison Between Blade-Element Models and Validation of a New Blade Element Momentum Skewed-Wake
of Propellers,” Aeronautical Journal, Vol. 112, No. 1138, Dec. 2008, Model within AeroDyn,” 33rd Wind Energy Symposium, AIAA Paper
pp. 689–704. 2015-0215, Jan. 2015.
https://doi.org/10.1017/s0001924000002669 https://doi.org/10.2514/6.2015-0215
[7] Bontempo, R., and Manna, M., “Analysis and Evaluation of the Momen- [26] Sun, Z., Shen, W. Z., Chen, J., and Zhu, W. J., “Improved Fixed Point
tum Theory Errors as Applied to Propellers,” AIAA Journal, Vol. 54, Iterative Method for Blade Element Momentum Computations,” Wind
No. 12, Dec. 2016, pp. 3840–3848. Energy, Vol. 20, No. 9, May 2017, pp. 1585–1600.
https://doi.org/10.2514/1.j055131 https://doi.org/10.1002/we.2110
[8] Bontempo, R., and Manna, M., “Effects of the Approximations Embod- [27] Masters, I., Chapman, J. C., Willis, M. R., and Orme, J. A. C., “A Robust
ied in the Momentum Theory as Applied to the NREL PHASE VI Wind Blade Element Momentum Theory Model for Tidal Stream Turbines
Turbine,” International Journal of Turbomachinery, Propulsion and Including Tip and Hub Loss Corrections,” Journal of Marine Engineer-
Power, Vol. 2, No. 2, June 2017, p. 9. ing & Technology, Vol. 10, No. 1, Jan. 2011, pp. 25–35.
https://doi.org/10.3390/ijtpp2020009 https://doi.org/10.1080/20464177.2011.11020241
[9] Bontempo, R., and Manna, M., “Highly Accurate Error Estimate of the [28] Ning, S. A., “A Simple Solution Method for the Blade Element Momen-
Momentum Theory as Applied to Wind Turbines,” Wind Energy, Vol. 20, tum Equations with Guaranteed Convergence,” Wind Energy, Vol. 17,
No. 8, March 2017, pp. 1405–1419. No. 9, June 2013, pp. 1327–1345.
https://doi.org/10.1002/we.2100 https://doi.org/10.1002/we.1636
[10] Bontempo, R., and Manna, M., “Verification of the Axial Momentum [29] Ning, A., “Using Blade Element Momentum Methods with Gradient-
Theory for Propellers with a Uniform Load Distribution,” International Based Design Optimization,” Structural and Multidisciplinary Optimi-
Journal of Turbomachinery, Propulsion and Power, Vo. 4, No. 2, May zation, Vol. 64, No. 2, May 2021, pp. 991–1014.
2019, p. 8. https://doi.org/10.1007/s00158-021-02883-6
https://doi.org/10.3390/ijtpp4020008 [30] Tollmien, W., Schlichting, H., Görtler, H., Tollmien, W., Schlichting, H.,
[11] Glauert, H., “Airplane Propellers,” Aerodynamic Theory, edited by W. F. Görtler, H., and Riegels, F. W., “Zusatz zu A. Betz, Schraubenpropeller
Durand, Springer, mit geringstem Energieverlust,” Ludwig Prandtl Gesammelte Abhand-
Berlin, 1935, pp. 169–360. lungen, Springer, Berlin, 1961, pp. 373–376.
https://doi.org/10.1007/978-3-642-91487-4_3 https://doi.org/10.1007/978-3-662-11836-8_27
[12] Shen, W. Z., Mikkelsen, R., Sørensen, J. N., and Bak, C., “Tip Loss [31] “Theory of Propellers. by Theodore Theodorsen. Published by
Corrections for Wind Turbine Computations,” Wind Energy, Vol. 8, McGraw-Hill & Co., 1948. 21/-net,” Journal of the Royal Aeronautical
No. 4, 2005, pp. 457–475. Society, Vol. 52, No. 453, Sept. 1948, pp. 574–574.
https://doi.org/10.1002/we.153 https://doi.org/10.1017/s0001924000098432
[13] Branlard, E., and Gaunaa, M., “Development of New Tip-Loss Correc- [32] Larrabee, E. E., “Practical Design of Minimum Induced Loss Propel-
tions Based on Vortex Theory and Vortex Methods,” Journal of Physics: lers,” SAE Technical Paper Series, SAE International, Warrendale, PA,
Conference Series, Vol. 555, Dec. 2014, Paper 012012. Feb. 1979, pp. 2053–2062.
https://doi.org/10.1088/1742-6596/555/1/012012 https://doi.org/10.4271/790585
[14] MacNeill, R., and Verstraete, D., “Blade Element Momentum Theory [33] Adkins, C. N., and Liebeck, R. H., “Design of Optimum Propellers,”
Extended to Model Low Reynolds Number Propeller Performance,” Journal of Propulsion and Power, Vol. 10, No. 5, Sept. 1994, pp. 676–682.
Aeronautical Journal, Vol. 121, No. 1240, May 2017, pp. 835–857. https://doi.org/10.2514/3.23779
https://doi.org/10.1017/aer.2017.32 [34] Wald, Q. R.. “The Aerodynamics of Propellers,” Progress in Aerospace
[15] Gur, O., and Rosen, A., “Propeller Performance at Low Advance Ratio,” Sciences, Vol. 42, No. 2, Feb. 2006, pp. 85–128.
Journal of Aircraft, Vol. 42, No. 2, March 2005, pp. 435–441. https://doi.org/10.1016/j.paerosci.2006.04.001
https://doi.org/10.2514/1.6564 [35] Hepperle, M., “Inverse Aerodynamic Design Procedure for Propellers
[16] Brandt, J., and Selig, M., “Propeller Performance Data at Low Reynolds Having a Prescribed Chord-Length Distribution,” Journal of Aircraft,
Numbers,” 49th AIAA Aerospace Sciences Meeting Including the New Vol. 47, No. 6, Nov. 2010, pp. 1867–1872.
Horizons Forum and Aerospace Exposition, AIAA Paper 2011-1255, https://doi.org/10.2514/1.46535
 RUH AND HWANG 4111

[36] Traub, L. W., “Inverse Propeller Design for a Prescribed Chord or Paper 102662.
Blade Angle,” Journal of Aircraft, Vol. 54, No. 2, March 2017, https://doi.org/10.1016/j.advengsoft.2019.03.005
pp. 825–830. [53] Drela, M., “XFOIL: An Analysis and Design System for Low Reynolds
https://doi.org/10.2514/1.c034023 Number Airfoils,” Lecture Notes in Engineering, Springer, Berlin, 1989,
[37] Traub, L. W., “Simplified Propeller Analysis and Design Including pp. 1–12.
Effects of Stall,” Aeronautical Journal, Vol. 120, No. 1227, May https://doi.org/10.1007/978-3-642-84010-4_1
2016, pp. 796–818. [54] Gelfand, I. M., and Silverman, R. A., Calculus of Variations, Courier
https://doi.org/10.1017/aer.2016.31 Corp., North Chelmsford, MA, 2000, Chap. 1.
[38] Traub, L. W., “Considerations in Optimal Propeller Design,” Journal of [55] Weinstock, R., Calculus of Variations: With Applications to Physics and
Aircraft, Vol. 58, No. 4, July 2021, pp. 950–957. Engineering, Courier Corp., New York, 1974, Chap. 4.
https://doi.org/10.2514/1.c036258 [56] Komzsik, L., Applied Calculus of Variations for Engineers, CRC Press,
[39] Davidson, R. E., “Optimization and Performance Calculation of Dual- Boca Raton, FL, Sept. 2018, Chap. 2.
Rotation Propellers,” NASA TP-1948, 1981. https://doi.org/10.1201/9781315215129
[40] Albert, S., Epple, P., Willinger, B., and Delgado, A., “High Efficiency [57] Ewing, G. M., and Hestenes, M. R., “Calculus of Variations and Optimal
Propeller Design Based on the Betz Minimum Induced Loss Condition Control Theory,” Mathematics of Computation, Vol. 21, No. 100,
and CFD Validation on an APC 8 Propeller,” Volume 7B: Fluids Oct. 1967, p. 739.
Engineering Systems and Technologies, American Soc. of Mechanical https://doi.org/10.2307/2005033
Engineers, Fairfield, NJ, Nov. 2013. [58] Gill, P. E., Murray, W., and Saunders, M. A., “SNOPT: An SQP
https://doi.org/10.1115/imece2013-64382 Algorithm for Large-Scale Constrained Optimization,” SIAM Review,
[41] Lock, C. N. H., Pankhurst, R. C., and Fowler, R. G., Determination of Vol. 47, No. 1, Jan. 2005, pp. 99–131.
the Optimum Twist of an Airscrew Blade by the Calculus of Variations, https://doi.org/10.1137/s0036144504446096
HM Stationery Office, London, U.K., 1942, pp. 1–27. [59] Polaczyk, N., Trombino, E., Wei, P., and Mitici, M., “A Review of
Downloaded by BEIHANG UNIVERSITY on February 3, 2024 | http://arc.aiaa.org | DOI: 10.2514/1.J062611

[42] Moriya, T., “Formulae for Propeller Characteristics Calculation and a Current Technology and Research in Urban On-Demand Air Mobility
Method to Obtain the Best Pitch Distribution,” Selected Scientific and Applications,” 8th Biennial Autonomous VTOL Technical Meeting and
Technical Papers, Univ. of Tokyo, Tokyo, 1938, pp. 43–47. 6th Annual Electric VTOL Symposium, Vertical Flight Soc., Mea,
[43] Dorfling, J., and Rokhsaz, K., “Constrained and Unconstrained Propel- Arizona, 2019, pp. 333–343.
ler Blade Optimization,” 52nd Aerospace Sciences Meeting, AIAA [60] Chauhan, S. S., and Martins, J. R., “Tilt-Wing eVTOL Takeoff Trajec-
Paper 2014-0563, Jan. 2014. tory Optimization,” Journal of Aircraft, Vol. 57, No. 1, Jan. 2020,
https://doi.org/10.2514/6.2014-0563 pp. 93–112.
[44] Chang, L. K., and Sullivan, J. P., “Optimization of Propeller Blade Twist https://doi.org/10.2514/1.c035476
by an Analytical Method,” AIAA Journal, Vol. 22, No. 2, Feb. 1984, [61] Gur, O., and Rosen, A., “Optimizing Electric Propulsion Systems for
pp. 252–255. Unmanned Aerial Vehicles,” Journal of Aircraft, Vol. 46, No. 4, July
https://doi.org/10.2514/3.48441 2009, pp. 1340–1353.
[46] Li, K., and Stefko, G., “Application of an Optimization Method to High https://doi.org/10.2514/1.41027
Performance Propeller Designs,” 20th Joint Propulsion Conference, [62] Gur, O., and Rosen, A., “Optimization of Propeller Based Propulsion
AIAA Paper 1984-1203, June 1984. System,” Journal of Aircraft, Vol. 46, No. 1, Jan. 2009, pp. 95–106.
https://doi.org/10.2514/6.1984-1203 https://doi.org/10.2514/1.36055
[47] Cho, J., and Lee, S. C., “Propeller Blade Shape Optimization for [63] Lee, J., and Hajela, P., “Parallel Genetic Algorithm Implementation in
Efficiency Improvement,” Computers & Fluids, Vol. 27, No. 3, Multidisciplinary Rotor Blade Design,” Journal of Aircraft, Vol. 33,
March 1998, pp. 407–419. No. 5, Sept. 1996, pp. 962–969.
https://doi.org/10.1016/s0045-7930(97)00035-2 https://doi.org/10.2514/3.47042.
[48] Burger, C., Hartfield, R., and Burkhalter, J., “Propeller Performance [64] Hoyos, J. D., Jiménez, J. H., Echavarría, C., Alvarado, J. P., and Urrea, G.,
Optimization Using Vortex Lattice Theory and a Genetic Algorithm,” “Aircraft Propeller Design through Constrained Aero-Structural Particle
44th AIAA Aerospace Sciences Meeting and Exhibit, AIAA Paper 2006- Swarm Optimization,” Aerospace, Vol. 9, No. 3, March 2022, p. 153.
1067, Jan. 2006. https://doi.org/10.3390/aerospace9030153
https://doi.org/10.2514/6.2006-1067 [65] Gray, J. S., Hwang, J. T., Martins, J. R., Moore, K. T., and Naylor, B. A.,
[49] Goldstein, S., “On the Vortex Theory of Screw Propellers,” Proceedings “OpenMDAO: An Open-Source Framework for Multidisciplinary
of the Royal Society of London. Series A, Containing Papers of a Design, Analysis, and Optimization,” Structural and Multidisciplinary
Mathematical and Physical Character, Vol. 123, No. 792, April 1929, Optimization, Vol. 59, No. 4, March 2019, Paper 10751104.
pp. 440–465. https://doi.org/10.1007/s00158-019-02211-z
https://doi.org/10.1098/rspa.1929.0078 [66] Hwang, J. T., and Ning, A., “Large-Scale Multidisciplinary Optimiza-
[50] Ruh, M. L., and Hwang, J. T., “Robust Modeling and Optimal Design of tion of an Electric Aircraft for On-Demand Mobility,” 2018 AIAA/
Rotors Using Blade Element Momentum Theory,” AIAA Aviation 2021 ASCE/AHS/ASC Structures, Structural Dynamics, and Materials
Forum, AIAA Paper 2021-2598, July 2021. Conference, AIAA Paper 2018-1384, Jan. 2018.
https://doi.org/10.2514/6.2021-2598 https://doi.org/10.2514/6.2018-1384
[51] Hwang, J. T., and Martins, J. R., “A Fast-Prediction Surrogate Model for [67] Ha, T. H., Lee, K., and Hwang, J. T., “Large-Scale Design-Economics
Large Datasets,” Aerospace Science and Technology, Vol. 75, April Optimization of eVTOL Concepts for Urban Air Mobility,” AIAA
2018, pp. 74–87. Scitech 2019 Forum, AIAA Paper 2019-1218, Jan. 2019.
https://doi.org/10.1016/j.ast.2017.12.030 https://doi.org/10.2514/6.2019-1218
[52] Bouhlel, M. A., Hwang, J. T., Bartoli, N., Lafage, R., Morlier, J., and
Martins, J. R., “A Python Surrogate Modeling Framework with J. Floryan
Derivatives,” Advances in Engineering Software, Vol. 135, Sept. 2019, Associate Editor