Professional Documents
Culture Documents
ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans
Research article
article info a b s t r a c t
Article history: The present paper is concerned with the dynamic modeling and design of control laws for a small
Received 16 June 2017 non-rigid multi-rotor airship constituted of an oblate-spheroid helium balloon coupled with an electric-
Received in revised form 22 January 2019 powered hexa-rotor airframe. The vehicle is assumed to operate in windless and low-speed conditions.
Accepted 8 March 2019
A six-degree-of-freedom nonlinear dynamic model is derived for it using the Newton–Euler approach
Available online xxxx
and considering, among other efforts, a restoring torque due to the displacement of the balloon’s
Keywords: center of buoyancy above the vehicle’s center of mass and the added-mass effect resulting from the
Multi-rotor airship air–structure interaction. Using the derived model and assuming a time-scale separation between the
Multi-rotor aerial vehicle translational and rotational dynamics, the attitude and position control laws are designed separately
Flight control system from each other. Both laws are formulated using feedback linearization combined with control input
Uncertainty quantification
saturation within appropriate parallelepipedal sets, which are carefully chosen to respect pre-defined
bounds on the control torque, control force and maximum inclination angle. The effect of temperature
and pressure fluctuations is taken into account through a parametric probabilistic approach, where
Maximum Entropy Principle is used to construct a physically consistent stochastic model and Monte
Carlo method is used as the stochastic solver to propagate the uncertainties through the system.
Extensive simulation results show the effectiveness of the proposed control system and quantify the
uncertainty of its performance over a wide range of local temperature and pressure.
© 2019 Published by Elsevier Ltd on behalf of ISA.
https://doi.org/10.1016/j.isatra.2019.03.010
0019-0578/© 2019 Published by Elsevier Ltd on behalf of ISA.
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
2 D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx 3
fi = kf ωi2 , (2)
τi = kτ ωi2 , (3)
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
4 D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx
B/G
where Ω B ∈ R3 is the SB representation of the vehicle’s angular
velocity w.r.t. SG .
Assume that the vehicle has a rigid structure and consider SG
as an inertial frame. Therefore, the Euler’s Law gives
B/G
[ ]
ḢB + Ω B × HB = TcB + TbB + TdB , (15)
tion of the (unknown) disturbance force. The force FbG is given = −Ma B/G −
g TaB Ω̇ B
by Eq. (7), while FG and FcG are modeled by
Ω BB/G × M11 vBB/G + M12 Ω BB/G
⎡ [ ]( ) ⎤
g ) ⎦, (18)
FG = − mt + mh ge3 ,
( ) ⎣ [
B/G
](
B/G B/G
) [
B/G
](
B/G B/G
(11) vB × M11 vB + M12 Ω B + Ω B × M21 vB + M22 Ω B
)T
FcG = F c DB/G e3 .
(
(12) where
[ ]
By replacing Eqs. (11)–(12) into (10), one can obtain M11 M12
Ma ≜ ∈ R6×6 , (19)
M21 M22
Fc Vg ρair
( )
B/G 1
r̈G = nG + − g e3 + FdG , (13)
mt + mh mt + mh mt + mh is the added mass in SB , Mij ∈ R3×3 , ∀i, j ∈ {1, 2}.
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx 5
By writing the added-mass force as well as the vehicle’s linear T̄cB ∈ R3 . The control allocation block is responsible for generating
velocity in SG , Eq. (18) yields the individual thrust commands f¯i ∈ R≥0 , i = 1, . . . , 6, from T̄cB
B/G B/G
[
B/G
] and F̄ c .
FaG = −DG/B M11 DB/G v̇G
− DG/B M12 Ω̇ B + DG/B M11 Ω B × This architecture is based on the assumption that there is a
B/G B/G B/G B/G time-scale separation between the (slower) closed-loop transla-
[ ]( )
× DB/G vG − DG/B Ω B × M11 DB/G vG + M12 Ω B , (20) tional and (faster) rotational vehicle dynamics [12]. This assump-
tion is ensured in practice by tuning the attitude control loop to
B/G B/G B/G
B/G
[ ]
converge much faster than the position control loop. Under such
TaB = −M21 DB/G v̇G + M21 Ω B × DB/G vG
− M22 Ω̇ B
assumptions, when designing the position control law, one can
B/G B/G B/G assume that the actual attitude DB/G converges to the correspond-
[( ) ]( )
− DB/G vG × M11 DB/G vG + M12 Ω B
ing command D̄B/G instantaneously, i.e., DB/G ≡ D̄B/G . On the
B/G B/G B/G
[ ]( )
− Ω B × M21 DB/G vG + M22 Ω B . (21) other hand, when designing the attitude control law, the attitude
command D̄B/G is assumed to be constant, or equivalently, the
B/G
Before computing Ma for the oblate-spheroid balloon under angular velocity command Ω̄ B ∈ R3 is assumed to be zero.
consideration, it is convenient to obtain the added mass M̌a ≜ On the basis of these conditions, the position and attitude control
diag(M̌11 , M̌22 ) with respect to a CCS which is parallel to SB and laws can be designed separately.
has its origin at CB, using the procedure of [21], resulting in
⎡ ⎤ 3.1. Position controller
m̌ 0 0 J̌ 0 0
[ ]
M̌11 = 0 m̌ 0 , and M̌22 =⎣ 0 J̌ 0 ⎦, (22)
First, the design model is obtained from Eq. (13), by assuming
0 0 m̌3 0 0 0 that: (1) the disturbance force FdG is negligible, (2) the actual
where control force magnitude F c is identical to the corresponding com-
4π α0 mand F̄ c , and (3) DB/G = D̄B/G (time-scale separation). The
m̌ ≜ a2 bρair , (23) resulting design model is
3 2 − α0
Vg ρair
( )
4π 2 β0 B/G 1
m̌3 ≜ a bρair , (24) r̈G = F̄c +
h G
− g e3 . (31)
3 2 − β0 mt + m mt + mh
4π 2 ) β0 − α0 Suppose that F̄cG is bounded within a parallelepipedal set from
a bρair a2 + b2 ,
(
J̌ ≜ (25)
15
√ 2 Fmin
≜ [F1min F2min F3min ]T ∈ R3 to Fmax ≜ [F1max F2max F3max ]T ∈ R3 .
1 − e2 ( −1 √ ) The position controller proposed here is the saturated law
α0 ≜ 3
sin e − e 1 − e2 , (26)
√e ( ) F̄cG = σ [Fmin ,Fmax ] (γ p ), (32)
2 1 − e2 e
β0 ≜ √ − sin e ,
−1
(27) [ p p p ]T
where γ p ≡ γ1 γ2 γ3 ∈ R3 is given by
e3 1 − e2
√ ) ( B/G B/G
)
a2 − b 2 γ p ≜ mt + mh g − Vg ρair e3 + mt + mh K1 r̄G − rG
(( ) ) (
e≜ 2
. (28)
a B/G
− mt + mh K2 ṙG ,
( )
(33)
Finally, to obtain Ma (in SB ), one can use the coordinate trans-
formation (see [22], p. 27) with the matrices K1 , K2 ∈ R3×3 representing the controller gains
(assumed to be diagonal), and
Ma = UT M̌a U (29)
σ[F min ,F max ] (γ1p )
⎡ ⎤
with 1 1
p ⎥
σ [Fmin ,Fmax ] (γ p ) ≜ ⎣ σ[F2min ,F2max ] (γ2 ) ⎦ , (34)
[ ] ⎢
I3 −d[e3 ×]
U≜ . (30) σ[F min ,F max ] (γ3p )
03×3 I3 3 3
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
6 D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx
Fig. 5. Constraint set of the position control input. (a) Perspective view showing the control force command inside the parallelepipedal set. (b) Superior view showing
the bounds F1max = F1max = F12max
. (c) Lateral view showing the maximum inclination angle ϕ max .
3.2. Attitude controller One can note that the proposed attitude control law (37)–
(38) is such that, if no saturation is active, it cancels the first
Here the design model is obtained from Eq. (17) by: (1) ne- three terms on the right-hand side of Eq. (38), remaining a
glecting the disturbance torque TdB ; (2) replacing the actual con- feedback-linearized closed-loop dynamic model controlled by the
trol torque TcB by the corresponding command T̄cB ; (3) considering proportional-derivative actions appearing in the last two terms of
that the rotor dynamics are so fast that one can assume ωi = ω̄i , (38). Section 3.5 presents a stability analysis.
∀i; and (4) assuming that ω̇i ≈ 0, ∀i. The resulting model is
[( 6
) ] 3.3. Attitude command generation
B/G B/G B/G
∑
Ω̇ =J −1
JΩ +J r
(−1) ω̄i e3
i
× Ω
In order to generate the three-DOF attitude command D̄B/R ,
B B B
i=1
note that, by definition, its third line is the transpose of n̄G . Then,
+ dF b J−1 [e3 ×]DB/G e3 + J−1 T̄cB . (36) consider the formula to convert from Euler angles (in the 1-2-3
sequence) to the attitude matrix (see [23], p. 52):
Suppose that the torque command T̄cB must be bounded from
−Tmax ∈ R3 to Tmax ≜ [T1max T2max T3max ]T . The attitude controller
⎡ ⎤
∗ ∗ ∗
B/R
proposed here is the saturated law D̄ =⎣ ∗ ∗ ∗ ⎦, (39)
sθ̄ −cθ̄ sφ̄ cθ̄ cφ̄
T̄cB = σ [−Tmax ,Tmax ] (γ ), a
(37)
where s and c stands for sin and cos, respectively, and compute
where
φ̄ and θ̄ from
B/G B/G
] [
γa ≜ −dF b [e3 ×]DB/G e3 + Ω B × JΩ B
φ̄ = −atan eT2 n̄G /eT3 n̄G , (40)
6
[
B/G
] ∑
B/G θ̄ = asin eT1 n̄G . (41)
+ Jr Ω B × e3 (−1)i ω̄i − JK3 ε − JK4 Ω B , (38)
i=1 Finally, considering the external heading command ψ̄ , one can
compute
with ε ∈ R containing the Euler angles (in the 1-2-3 sequence)
3
)T
corresponding to the attitude control error D̃ = DB/G D̄B/G , cψ̄ cθ̄ cψ̄ sθ̄ sφ̄ + sψ̄ cφ̄ −cψ̄ sθ̄ cφ̄ + sψ̄ sφ̄
( ⎡ ⎤
K3 , K4 ∈ R3×3 are controller (diagonal-matrix) gains, and the D̄B/R = ⎣ −sψ cθ̄ −sψ̄ sθ̄ sφ̄ + cψ̄ cφ̄ sψ̄ sθ̄ cφ̄ + cψ̄ sφ̄ ⎦ . (42)
saturation function σ [−Tmax ,Tmax ] (.) is as defined in Eqs. (34)–(35). sθ̄ −cθ̄ sφ̄ cθ̄ cφ̄
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx 7
3.4. Control allocation globally exponentially stable. This implies that, for any λ ∈ R>0 ,
˙
if (r̃(0), r̃(0)) ˙
∈ Bλ , then (r̃(t), r̃(t)) ∈ Bλ , ∀t ∈ R>0 , which means
Consider Eq. (5) again; it relates the effective resultant efforts that Bλ is an invariant set of system (48).
F c and TcB with the six individual rotor forces (stored in f). We can In particular, consider the ball L∗p defined in (47). Since, by
consider that the respective effort commands are interrelated in construction, L∗p ⊆ Lp , it holds that γ p ∈ F , from the above
the same way, i.e., argumentation, one can see that L∗p is an invariant set of system
B/G
(31)–(33), which, in this case, behaves as (48) (with r̄¨G (t) =
[ ]
F̄ c
= Γ f̄, (43) ˙
0, ∀t ∈ R≥0 ). Therefore, if (r̃(0), r̃(0)) ˙
∈ L∗p , then (r̃(t), r̃(t)) ∈
T̄cB
]T Lp , ∀t ∈ R>0 and r̃ converges to 03×1 exponentially. From the
∗
B/G B/G
where f̄ ≜ f¯1 f¯2 . . . f¯6 .
[
definition of r̃, this is equivalent to say that (rG (t), ṙG (t)) ∈
B/G B/G B/G B/G
The control allocation pursued here is to find a solution f̄ to the L∗p ⊕ (r̄G (t), r̄˙ G (t)), ∀t ∈ R≥0 and rG converges to r̄G
system of linear equations (43). Since this is an undetermined sys- exponentially, thus completing the proof. □
tem (with infinitely many solutions), we choose here the one that
minimizes ∥f̄∥2 , considering (43) as a set of equality constraints. 3.5.2. Attitude control
The explicit solution to such a problem is well known; by using Denote the admissible-torque set
Lagrange multipliers, it can be found to be
T ≜ T ∈ R3 : −Tmax ≤ T ≤ Tmax
{ }
[ ] (49)
c
)−1 F̄
f̄ = Γ T Γ Γ T .
(
(44) and define the corresponding viable set
T̄cB
B/G
{( ) }
The above control allocation can be seen as an energy-efficient La ≜ ε, Ω B ∈ R6 : γ a ∈ T , (50)
solution, since it minimizes the squared norm of f̄. Moreover, con-
sidering that the commands F̄ c and T̄cB belong to the admissible as well as the maximal 0-centered Euclidean ball inside La
set of the above optimization problem (meaning that the problem L∗a ≜ max Bλ s.t . Bλ ⊆ La . (51)
has at least one solution), the resulting control allocation is exact, λ
i.e., it satisfies Eq. (43). From the assumption that D̄B/G (t) is constant (time-scale sepa-
B/G
ration), it follows that Ω̄ B (t) = 0. In this case, the attitude error
3.5. Stability analysis
kinematics can be described by (see [23], p. 73)
B/G
Still considering the time-scale separation assumption, this ε̇ = A(ε)Ω B , (52)
subsection presents stability proofs for both the position and
attitude control loops. where
cε3 /cε2 −sε3 /cε2 0
[ ]
3.5.1. Position control A(ε) ≜ s ε3 cε3 0 , (53)
Denote the admissible-force set by −cε3 sε2 /cε2 sε3 sε2 /cε2 1
where Bλ ≜ x ∈ R6 : ∥x∥ ≤ λ . The following theorem gives a stability result for the closed-
{ }
The following theorem gives a stability result for the closed- loop system (52), (37)–(38).
loop system (31)–(33).
Theorem 2. Consider the closed-loop system (52), (37)–(38). As-
Theorem 1. Consider the closed-loop system (31)–(33). Assume that sume that D̄B/G (t) is constant at any t ∈ R≥0 (time-scale separation)
B/G B/G
the position command r̄G is such that r̄¨G (t) = 0, ∀t ∈ R≥0 . In this and that ε is so small that its kinematics can be described by (54).
B/G
B/G B/G B/G B/G In this context, if the initial state (ε(0), Ω B (0)) ∈ L∗a , then DB/G
context, if the initial state (rG (0), ṙG (0)) ∈ L∗p ⊕ (r̄G (0), r̄˙ G (0)), B/G B/G
B/G B/G B/G B/G converges exponentially to D̄ and (ε(t), Ω B (t)) ∈ L∗a , ∀t ∈ R>0 .
then rG converges exponentially to r̄G and (rG (t), ṙG (t)) ∈
B/G B/G
L∗p ⊕ (r̄G (t), r̄˙ G (t)), ∀t ∈ R>0 .
Proof. Under the condition that γ a ∈ T or, equivalently,
B/G
(ε, Ω B ) ∈ La , and considering the small-angle kinematics de-
Proof. Under the condition that γ p ∈ F or, equivalently, scribed by (54), the closed-loop rotational dynamics represented
˙ ∈ Lp , the closed-loop translational dynamics represented
(r̃, r̃) by Eqs. (54),(37)–(38) can be re-modeled in terms of ε and its
by Eqs. (31)–(33) can be re-modeled in terms of r̃ and its time time derivatives, resulting in
derivatives, resulting in
B/G
ε̈ + K4 ε̇ + K3 ε = 0. (55)
r̃¨ + K2 r̃˙ + K1 r̃ = −r̄¨G . (48)
Since the gain matrices K3 and K4 are positive-definite and
B/G
Since, by assumption, r̄¨G (t) = 0, ∀t ∈ R≥0 and the gain diagonal, from the linear control theory, one see that the state
matrices K3 and K4 are positive-definite and diagonal, from the (ε, ε̇) = (03×1 , 03×1 ) of system (55), without considering the
˙ = (03×1 , 03×1 )
linear control theory, one see that the state (r̃, r̃) control saturation, is globally exponentially stable. This implies
of system (48), without considering the control saturation, is that, for any λ ∈ R>0 , if (ε(0), ε̇(0)) ∈ Bλ , then (ε(t), ε̇(t)) ∈
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
8 D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx
Fig. 6. Nominal performance of the position control. It shows the controlled position components together with the respective command.
Fig. 7. Nominal performance of the attitude control. It shows the controlled attitude variables together with the respective command.
Fig. 8. Force magnitude command provided by the position controller in the Fig. 9. Torque command provided by the attitude controller in the nominal
nominal simulation. simulation.
Bλ , ∀t ∈ R>0 , which means that Bλ is an invariant set of system It is worth noting at this point that both the aerostatic lift
(55). F⃗ b and the restoring torque T⃗ b depend on the value of ρair and
In particular, consider the ball L∗a defined in (51). Since, by ρhelium . These densities in turn depend on the local temperature
construction, L∗a ⊆ La , it holds that γ a ∈ T , from the above and pressure, as expressed by the Ideal Gas Law:
argumentation and assumptions, one can see that L∗a is an in-
variant set of system (54), (37)–(38), which, in this case, be- p
haves as (55). Therefore, if (ε(0), ε̇(0)) ∈ L∗a , then (ε(t), ε̇(t)) ∈ ρq = , (56)
Rq T
L∗a , ∀t ∈ R>0 and ε converges to 03×1 exponentially. From the
definition of ε and considering (54), this is equivalent to say
B/G where ‘‘q" can be replaced by either ‘‘air’’ or ‘‘helium’’, p is the
that (DB/G ; Ω B ) converges to (D̄B/G ; 03×1 ) exponentially, thus local pressure in Pa, T is the local temperature in K, and Rq is
completing the proof. □ the specific gas constant (Rair = 286.9 J/Kg/K and Rhelium =
2077 J/Kg/K). In particular, in this section, the surrounding air
and contained gas are supposed to have the same temperature
4. Nominal simulation
and pressure; their values are 20o C and 101325 Pa, respectively.
Moreover, the air is assumed to be dry. Under such conditions,
As a first step to evaluate the proposed method, this section the air and helium densities are ρair ≈ 1.204 kg/m3 and ρhelium ≈
presents a numerical simulation of the proposed flight control 0.1664 kg/m3 . These values are used here to simulate the ground
system in nominal conditions. truth as well as to compute the control laws.
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx 9
Section 4.1 describes the simulation and shows the adopted ambient the temperature and pressure are modeled as indepen-
nominal parameters, while Section 4.2 presents and analyzes the dent random variables and, based on such probabilistic models,
simulation results. for each flight realization, one new outcome is generated for
both variables. For simulating the plant, the surrounding air and
4.1. Plant and controller parameters contained helium are assumed to have the same (varying) tem-
perature and pressure. On the other hand, in the implementation
For simulating the overall closed-loop flight control system of the control laws, their densities are maintained constant (and
represented in Fig. 4, we use the models formulated in Sec- equal the respective nominal values; see Section 4).
tion 2 as well as the control laws and control allocation proposed
in Section 3. The simulator is implemented in MATLAB, using 5.1. Probabilistic model
the fourth-order Runge–Kutta solver, with an integration step
of 0.001 s. Table 1 shows the nominal values of the hexa-rotor Let (Θ , Σ, P) be a probability space, where Θ is the sample
airship parameters. On the other hand, Table 2 presents the space, Σ is a σ -algebra over Θ , and P : Σ → [0, 1] a probability
adopted controller parameters. The controller gains are tuned by measure. In this stochastic framework, the local temperature T l
trial and error, taking into account their proportional or derivative and pressure P l are modeled by (independent) random variables
effect and considering the time-scale separation assumption as X : Σ → R and Y : Σ → R, respectively. We assume that the joint
well. probability distribution of these random variables, PXY (dx dy),
In this paper, for obtaining simulation data that are consistent admit a probability density function (PDF) (x, y) ↦ → pXY (x, y),
with a typical operation of MAVs, the proposed flight control with respect to dx dy. Similarly, the marginal distributions PX (dx)
system is commanded to follow a waypoint-based position tra- and PY (dy) are assumed as having marginal PDFs x ↦ → pX (x), with
jectory. In this trajectory, the waypoints are connected by straight respect to dx, and y ↦ → pY (y), with respect to dy.
lines with length of 5 m and a constant desired velocity of v̄ = The maximum entropy principle [26] is adopted here as a way
0.5 m/s. Moreover, the heading angle command ψ̄ is set to zero. to specify the aforementioned distributions. This approach pro-
vides the least biased distributions in a scenario with little (or no)
4.2. Nominal simulation results experimental information on a particular random variable [24].
In the present investigation, the random local temperature X and
Figs. 6–8 are the results of the nominal simulation using the local pressure Y are both completely unknown beforehand. How-
parameters of Tables 1–2. Fig. 6 shows the effective position and ever, one should better specify for them reasonable operational
the corresponding position command. In the ramp parts of the ranges, from which it is possible to immediately define the under-
component trajectories, one can verify a steady-state error of lying supports. In this way, we assume from now on that X and Y
about 2 m. After finishing the ramps, there are no overshoots take values in known positive finite intervals [x1 , x2 ] ⊂ (0, +∞)
and the accommodation times (within 5 cm of the final value) and [y1 , y2 ] ⊂ (0, +∞), respectively. Therefore, the maximum
are between 10 s and 14 s. In the flat parts of the trajectory, entropy principle is formulated here as the maximization of the
the steady-state errors are zero. The large errors during the Shannon entropy
ramps and the long accommodation times reflect the vehicle’s
∫ ∫
inherent low acceleration capability, which can be attributed to S (pXY ) = − ln pXY (x, y) pXY (x, y) dx dy , (57)
R2
both the restricted inclination bound ϕ max and the large inertia
parameters. subject to the normalization constraint (which brings information
Fig. 7 shows the effective and commanded attitude along the about the supports)
time using Euler angles in the 1-2-3 sequence. The attitude pitch ∫ y2 ∫ x2
and roll commands deviate from zero at the waypoints and the pXY (x, y) dx dy = 1. (58)
corresponding effective angles follow them with an apparently y1 x1
fast but attenuated response. In the yaw angle, one can see a By using the Lagrange multiplier method, one can solve the
very small and short transient starting at 20 s; it is due to the above optimization problem to obtain the following PDF
gyroscopic coupling of the rotational dynamics, which becomes
apparent when there exist a simultaneous motion in roll and pXY (x, y) = pX (x) × pY (y), (59)
pitch.
with
Finally, Fig. 8 presents the force magnitude command F̄ c com-
puted by the position controller and Fig. 9 shows the components 1 1
pX (x) = 1[x1 ,x2 ] (x) and pY (y) = 1[y1 ,y2 ] (y) , (60)
of the torque command T̄cB generated by the attitude controller. x2 − x1 y2 − y1
There are transients at the instants the vehicle is commanded where 1X (x) denotes the indicator function of the set X . Note
to ascend or to stop ascending, after which, F̄ c converges and that the above marginal PDFs correspond to uniform distribu-
stays around 42.1 N. This force value is equivalent to the total tions on the intervals [x1 , x2 ] and [y1 , y2 ], respectively. Informally,
weight of the vehicle, payload, and helium, minus the aerostatic the above result makes all sense, because since the minimum
lift. The force command F̄ c keeps far from its maximum bound, and maximum values are the only known information about the
even during the ascending acceleration. On the other hand, the random parameters, the most reasonable choice for their distri-
components of the torque command T̄cB deviate from zero at butions is the one that assigns equal weight to all their possible
the waypoints, but stay inside their bounds ±Tmax with a large values.
margin. Due to the randomness of X and Y, the air density ρair and the
helium density ρhelium are also random variables [see Eq. (56)].
5. Uncertainty quantification Further, since the aerostatic lift [see Eq. (7)] and the restoring
torque [see Eq. (9)] depend on ρair and ρhelium , then the trans-
The present section conducts an uncertainty quantification lational (14) and rotational (17) dynamic models are rather dif-
analysis of the flight control system under consideration us- ferential equations with random parameters and, consequently,
ing a parametric probabilistic approach [24,25]. In this analysis, their variables are stochastic processes. In particular, we choose
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
10 D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx
Table 1
Physical parameters of the hexa-rotor airship.
Description Symbol Value
Force coefficient kf 1.2838 × 10−5 N/(rad/s)2
Torque coefficient kτ 3.0811 × 10−7 N m/(rad/s)2
Maximum rotor speed ω̄max 906.66 rad/s
Motor speed coefficient kω 1
Motor time constant τω 0.01 s
Arm length l 1 m
Volume of the balloon V 5.3 m3
CB–CM displacement d 0.85 m
Semi-major axis a 1.25 m
Semi-minor axis b 0.8 m
Total inertia matrix J diag(2.0633, 2.0651, 1.9556) kg m2
Moment of inertia of the rotors Jr 0.001 kg m2
Total empty mass mt 9.392 kg
Table 2
Parameters of the attitude and position control laws.
Description Symbol Value
Proportional gain of the position controller K1 diag(0.5, 0.2, 0.7)
Derivative gain of the position controller K2 diag(2, 1, 3)
Proportional gain of the attitude controller K3 diag(20, 50, 1)
Derivative gain of the attitude controller K4 diag(10, 20, 1)
]T
Tmax 16.3 14.1 0.58 Nm
[
Maximum torque command
]T
min
−5.8 −5.8 2.7 N
[
Minimum force command F
]T
Fmax 5.8 5 .8 54.6 N
[
Maximum force command
Maximum inclination angle ϕ max 12 degrees
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx 11
Fig. 11. Position statistics (mean and confidence region) estimated by Monte Carlo method.
Fig. 12. Attitude control error statistics (mean and confidence region) estimated by Monte Carlo method.
Fig. 13. Torque command statistics (mean and confidence region) estimated by Monte Carlo method.
Fig. 14. Force magnitude command statistics of the Monte Carlo simulation. Fig. 15. Aerostatic lift statistics of the Monte Carlo simulation.
6. Concluding remarks structurally simple concept able to extend the flight duration
and load capacity of multi-rotor vehicles. The operation scenarios
The paper was concerned with the dynamic modeling and taken into account involve low-speed and windless conditions,
flight control of a hybrid airship consisting of a hexa-rotor where the most relevant aerodynamic effects on the airship
airframe attached to an oblate-spheroid helium balloon. It is a motion are due to the so-called added mass.
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.
12 D.A. dos Santos and A. Cunha, Jr. / ISA Transactions xxx (xxxx) xxx
The proposed flight control scheme was numerically evaluated [8] Frye MT, Gammon SM, Qian C. The 6-DOF dynamic model and simulation
in two phases. First, considering constant and known nominal of the tri-turbofan remote-controlled airship. In: Proceedings of the 2007
american control conference. New York; 2007.
parameters, the simulation results showed that, in low speed, it
[9] Azouz N, Chaabani S, Lerbet J, Abichou A. Computation of the added masses
is possible to control the vehicle to track a sequence of waypoints of an unconventional airship. J Appl Math 2012;2012:1–19.
without position overshoots or attitude oscillations. Second, using [10] Lanteigne E, Alsayed A, Robillard D, Recoskie SG. Modeling and con-
Monte Carlo method, an uncertainty quantification analysis was trol of an unmanned airship with sliding ballast. J Intell Robot Syst
carried out to assess the sensibility of the overall system with 2017;88:285–97.
[11] Araujo LS. Unmanned ellipsoid multi-rotor airship and respective method
respect to an unknown local temperature and pressure condition of construction. US patent, number 2018/0319476 A1; 2018.
(which affects the air and helium densities and, consequently, the [12] Bertrand S, Guénard N, Hamel T, Piet-Lahanier H, Eck L. A hierarchical
aerostatic efforts). It showed that the consequent performance controller for miniature VTOL UAVs: Design and stability analysis using
variation in the rotational and horizontal–translational motion is singular perturbation theory. Control Eng Pract 2011;19(10):1099–108.
http://dx.doi.org/10.1016/j.conengprac.2011.05.008,
small; the only relevant variation occurs in the altitude motion.
[13] Lee T, Leok M, McClamroch NH. Nonlinear robust tracking control of
It revels that the air and helium densities used to parameterize a quadrotor UAV on SE(3). Asian J Control 2013;15(2):391–408. http:
the control laws must be computed before the flight, considering //dx.doi.org/10.1002/asjc.567.
the actual temperature and pressure condition. [14] Shi X-N, Zhang Y-A, Zhou D. Almost-global finite-time trajectory tracking
control for quadrotors in the exponential coordinates. IEEE Trans Aerosp
Electron Syst 2017;PP(99). http://dx.doi.org/10.1109/TAES.2017.2649258.
Acknowledgments [15] Santos DA, CunhaJr A. Dynamic modeling and flight control of a balloon-
quadcopter unmanned aerial vehicle. In: Proceedings of the XXXVIII
This work received partial support from Elio Technologia Ltda, iberian latin-american congress on computational methods in engineering.
São Paulo Research Foundation (FAPESP) (grant number Florianopolis; 2017.
[16] Tuveri M, Ceruti A, Marzocca P. Added masses computation for uncon-
2017/06877-2), and Carlos Chagas Filho Research
ventional airships and aerostats through geometric shape evaluation and
Foundation of Rio de Janeiro State (FAPERJ) (grants number E- meshing. Int J Aeronaut Space Sci 2014;15(3):241–57.
26/010.000.805/2018). [17] Mahony R, Kumar V, Corke P. Multirotor aerial vehicles: Modeling, estima-
tion, and control of quadrotor. IEEE Robot Autom Mag 2012;19(3):20–32.
Conflict of interest http://dx.doi.org/10.1109/MRA.2012.2206474,
[18] Fossen TI. Guidance and control of ocean vehicles. New York: Wiley; 1998,
p. 5–42.
The authors declare that they have no known competing finan- [19] Thomasson PG. Equations of motion of a vehicle in a moving fluid. J Aircr
cial interests or personal relationships that could have appeared 2000;37(4).
to influence the work reported in this paper. [20] Li Y, Nahon M. Modeling and simulation of airship dynamics. J Guid Control
Dyn 2007;30(6):1691–700.
[21] Tuckerman IB. Inertia factors for ellipsoids for use in airship design.
References National Advisory Committee for Aeronautics; 1925.
[22] Li Y. Dynamics modeling and simulation of flexible airships [Ph.D. thesis],
[1] Khoury AG, editor. Airship technology. Cambridge University Press; 2012. McGill University; 2008.
[2] Elfes A, Bueno SS, Mergerman M, Ramos JG. A semi-autonomous robotic [23] Markley FL, Crassidis JL. Fundamentals of spacecraft attitude determination
airship for environmental monitoring missions. Leuven; 1998. p. 3449–55. and control. Springer; 2014.
[3] Gomes SBV, Ramos JG. Airship dynamic modeling for autonomous opera- [24] Soize C. Stochastic modeling of uncertainties in computational structural-
tion. In: Proceedings of the 1998 IEEE international conference on robotics dynamics - Recent theoretical advances. J Sound Vib 2013;332(10):2379–
and automation. Leuven; 1998. p. 3462–7. 95. http://dx.doi.org/10.1016/j.jsv.2011.10.010.
[4] Li Y, Nahon M, Sharf I. Airship dynamics modeling: A literature review. [25] Soize C. Uncertainty quantification: An accelerated course with advanced
Prog Aerosp Sci 2011;47:217–39. applications in computational engineering. Springer; 2017.
[5] Moutinho A, Azinheira JR. Stability and robustness analysis of the AURORA [26] Jaynes ET. Information theory and statistical mechanics. Phys Rev Ser II
airship control system using dynamic inversion. In: Proceedings of the 1957;106:620–30. http://dx.doi.org/10.1103/PhysRev.106.620.
2005 IEEE international conference on robotics and automation. Barcelona; [27] Rubinstein RY, Kroese DP. Simulation and the Monte Carlo method, 3rd
2005. p. 2265–70. ed. Wiley; 2016.
[6] Liesk T, Nahon M, Boulet B. Design and experimental validation of a [28] Soize C. A comprehensive overview of a non-parametric probabilistic ap-
nonlinear low-level controller for an unmanned fin-less airship. IEEE Trans proach of model uncertainties for predictive models in structural dynamics.
Control Syst Technol 2013;21(1):147–61. J Sound Vib 2005;288:623–52. http://dx.doi.org/10.1016/j.jsv.2005.07.009.
[7] Earon E, Rabbath CA, Apkarian J. Design and control of a novel hybrid
vehicle concept. In: AIAA guidance, navigation, and control conference and
exhibit. Hilton Head, South Carolina; 2007. p. 1–8.
Please cite this article as: D.A. dos Santos and A. Cunha, Jr., Flight control of a hexa-rotor airship: Uncertainty quantification for a range of temperature and pressure
conditions. ISA Transactions (2019), https://doi.org/10.1016/j.isatra.2019.03.010.