Professional Documents
Culture Documents
net/publication/330997554
CITATIONS READS
4 2,569
4 authors, including:
Al Al
ITP Padang
8 PUBLICATIONS 18 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
For Quadcopter Capability Development Additional Low Voltage Distribution Network Location Tracking View project
a r t i c l e i n f o a b s t r a c t
Article history: A piloted simulator evaluation is an important step for making a Fault-Tolerant Flight Control (FTFC)
Received 2 July 2019 system more practical and reliable. Especially, in the case of an adaptive technique, conformance of the
Received in revised form 29 May 2020 adaptive controller to the human pilot’s intended maneuver must be investigated carefully since the
Accepted 4 August 2020
overall control performance will be changed by the additional loop closed by the pilot. In this paper,
Available online 10 August 2020
Communicated by Huihe Qiu
we present a pilot-in-the-loop simulation of Simple Adaptive Control (SAC) with a PID compensator. By
SAC adjusting to the faulty aircraft dynamics, the proposed FTFC system reduces the pilot workload in
Keywords: emergency situations.
Fault-tolerant control © 2020 The Authors. Published by Elsevier Masson SAS. This is an open access article under the CC BY
Simple adaptive control license (http://creativecommons.org/licenses/by/4.0/).
Pilot-in-the-loop simulation
https://doi.org/10.1016/j.ast.2020.106125
1270-9638/© 2020 The Authors. Published by Elsevier Masson SAS. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
2 R. Takase et al. / Aerospace Science and Technology 106 (2020) 106125
are calculated from the pilot’s control inputs. These ideal states are and y (t ) ∈ Rm is the measurement output. In addition, Eqs. (4) and
then used as reference values, to be tracked as closely as possible (5) are supposed to satisfy the Almost Strictly Positive Real (ASPR)
by the damaged aircraft using the PID-SAC system. This way, pi- condition.
lots will feel as if they are controlling the ordinary (undamaged) In this paper, we define the following linear system as the so
aircraft. called ‘reference model’ G m (s), which is not to be confused with
the ‘ideal model’ discussed in the previous subsection.
2.2. Ideal model
ẋm (t ) = A m xm (t ) + B m um (t ) (6)
Although a nonlinear aircraft dynamics model (e.g. the actual ym (t ) = C m xm (t ) (7)
plant model used in the simulation) might seem an obvious choice
for the ideal model, any model generating a meaningful reference where xm (t ) ∈ R , um (t ) ∈ R , and ym (t ) ∈ R , and we assume
nm m m
signal can be considered. In this paper, we choose a linearized nm ≤ n. Even if the parameters of the system in Eqs. (4) and (5)
model to illustrate that the ideal model and the plant do not are unknown, we can find a control input u (t ) which drives the
need to be the same, while not unnecessarily complicating the pi- plant output y (t ) to the reference model output ym (t ), only when
lot’s control task. In addition, FTFC should in particular support the ASPR condition is satisfied [17]. In this case, the control input
the pilot in keeping the steady-state flight condition, where the is given by
linearized model provides a good approximation of the nonlinear
u (t ) = K (t ) z(t ) (8)
dynamics. Thus, we use the following linearized model as the ideal
T T T T
model: z(t ) = [e (t ) xm (t ) um (t ) ] (9)
K (t ) = [ke (t ) kxm (t ) kum (t )] (10)
ẋi (t ) = A i xi (t ) + B i u i (t ) (1)
⎡ ⎤ ⎡ ⎤ e (t ) = y (t ) − ym (t ) (11)
Yv W0 + Y p −U 0 + Y r g cos θ0 Y δa Y δr
⎢ Lv Lp Lr 0 ⎥ ⎢ Lδ L δr ⎥ and shown in the block diagram in Fig. 2. Since the typical aircraft
Ai = ⎢
⎣ Nv
⎥ Bi = ⎢ a
⎦ ⎣ N δa
⎥
Np Nr 0 N δr ⎦ dynamics equations are not ASPR, a parallel feedforward compen-
0 1 tan θ0 0 0 0 sator (PFC) is added. Details can be found in [17].
R. Takase et al. / Aerospace Science and Technology 106 (2020) 106125 3
Table 1
Parameters of the pilot model.
Roll Sideslip
K Pa -0.20 K Pr 0.10
T La 0.01 T Lr 0.001
T Ia 0.1 T Ir 0.01
Table 2
Assumed scenarios for the piloted evaluation.
Fault No. Scenario characteristics
0 No fault (before 80 s)
1 only 20% of aileron
effectiveness after 80 s
2 only 20% of aileron & rudder
Fig. 2. The structure of SAC system with PID compensator.
effectiveness after 80 s
We use the following adjustment rule for adapting the control where the parameters of the PID compensator were set so as to
gains [18]. have high target tracking performance before the fault occurs. The
PID gains of the aileron input are ( K P , K I , K D ) = (3.0, 2.07, 0.5),
K̇ (t ) = −e (t ) z(t ) T I − σ K (t ) (12) and PID gains of the rudder input are ( K P , K I , K D ) = (8.0, 1.3,
0.03).
where σ is a constant in order to avoid the burst phenomena, and The PFC used in this paper was designed via Iterative Linear
I ∈ R (nm +2m)×(nm +2m) is an adaptation rate. A detailed investiga-
Matrix Inequalities (ILMI) following Ref. [10]. The idea of design-
tion of SAC algorithms and their control performance can be found ing a PFC and the details of the ILMI algorithm are described in
in e.g. [19,20]. Ref. [8] and Ref. [21], respectively. Note that the PFC is designed
only for the nominal plant model, which means that the ASPR
2.4. Application of SAC to flight control condition may not be satisfied after the faults. Therefore, after de-
signing the PFC for the nominal model, it has been confirmed that
The authors’ group has applied SAC with PID control to MIMO the PFC works appropriately for both the nominal and fault condi-
flight dynamics and the flight experiment results have been re- tions through a ground and an actual flight test [10].
ported [10]. In Ref. [10], adaptive gains were updated for roll and
yaw angle simultaneously. However, this method may not be ap- 3. Simulation with pilot model
propriate for the fault tolerant flight controller considering pilot
inputs. The reason is explained as follows. Before the human-in-the-loop experiments, the basic control
In the case of pilot manual control, the roll and sideslip an- performance is investigated through a series of simulations with
gle are mainly controlled by the aileron and rudder respectively. a pilot model. In this study, we use the following pilot model [22].
In addition, pilots firstly try to control the roll angle since the
TLs + 1 τs
roll angle is the innermost loop in the flight dynamics. These Y p (s) = K p e (19)
facts suggest that adaptive gains of aileron inputs should be up- TIs + 1
dated for only desired roll angle. This means that adaptive gains where K P , T L , T I , and τ are the pilot static gain, the lead time
of aileron and rudder inputs should be independently updated for constant, the lag time constant, and the effect of time delay, re-
roll and sideslip angle even in the case of MIMO lateral motion. spectively. The parameters are tuned to satisfy the desired tracking
Thus, the structure of SAC is a combination of two SISO systems, performance in the fault free case, and are summarized in Ta-
which means that in the case of the roll angle, SAC is applied ble 1.
for the transfer function from aileron input to roll angle, while The simulation is carried out in a MATLAB/Simulink environ-
another SAC takes care of the rudder to sideslip angle transfer ment around a trim condition. The aircraft dynamics model is
function. representing JAXA’s Multi-Purpose Aviation Laboratory (Mupal-α ),
The reference models and adjustment parameters of the roll a modified Do228-200. For the simulation, the linearized aircraft
and sideslip angle are the same, and as follows. model is used, whose trim condition is at an altitude of 1,520
m and an airspeed of 66.8 m/s [23]. We assume two fault cases
0.05 · ẋm (t ) = −xm (t ) + um (t ) (13) (Table 2). The first is that the ailerons’ effectiveness reduces to
20% of their ordinary state at 80 s. The second is that both the
ym (t ) = xm (t ) (14) aileron and rudder effectiveness reduce to 20% of their ordinary
1 states at 80 s. The reductions in aileron and rudder effectiveness
G m (s) = (15) are emulated by reducing the control gains and limits in the sim-
0.05s + 1
ulation.
I = diag(3, 0.01, 0.01) (16) Fig. 3 shows the time histories of roll angle in the upper fig-
ure and that of sideslip angle in the lower one for the case of the
σ = 0.01 (17) Fault No. 2. The solid line denoted by “Pilot model with FTFC” is
the behavior of the pilot model with PID-SAC. The dash-dot line
where xm (t ) ∈ R, um (t ) ∈ R, and G m (s) is the transfer function of
denoted by “Pilot model” is the behavior of only the pilot model.
the reference model.
The dotted line denoted by “Pilot model (switched)” is the behav-
The PID compensator C P I D (s) is described as
ior of the pilot model whose gain becomes five times the original
1 value after the fault, reflecting the “ideal” way of coping with the
C P I D (s) = K P + K I + KDs (18) reduced control surface effectiveness. After the fault occurs on the
s
4 R. Takase et al. / Aerospace Science and Technology 106 (2020) 106125
Fig. 3. Comparison of roll and sideslip angle time histories with and without the Fig. 4. Time histories of adaptive gains for aileron command (Fault No. 2).
proposed PID-SAC FTFC when the system is controlled by a pilot model (Fault No. 2).
(For interpretation of the colors in the figure(s), the reader is referred to the web
version of this article.)
1
N
AVGdefl = |δctrl (k)| (20)
N
k =1
Fig. 5. Time histories of adaptive gains for rudder command (Fault No. 2).
where δctrl (t ) is the pilot control input and N is the number of
recorded data samples. In order to compare between the no fault
case and the fault cases, the values of the average absolute de-
flection are normalized by those obtained in the no fault case
(Fault No. 0). Fig. 6 shows the normalized average absolute aileron
and rudder deflections. In the case that the pilot model is ideally
switched after the fault, the normalized average absolute deflec-
tions become larger due to the fault. Note that the workloads
become five times as large as comparing similar sections before
and after the fault, e.g., the section of [20, 80] s and [100, 160] s.
This means that in the case of a human pilot, he has to adjust to
the faulty dynamics, so the pilot workload will be larger after the
fault. On the other hand, the FTFC can track the desired value af-
ter the fault by automatically adjusting the adaptive gains (Figs. 4
and 5). This means that the FTFC adjusts to the faults instead of
the pilot model. From this simulation, the basic performance of
the proposed FTFC system has been shown for the pilot model. In
the next section, we will explain the human-pilot-in-the-loop sim-
ulation result.
Fig. 6. Average absolute deflection of control surface normalized by the no-fault case
4. Human-piloted simulation deflection when the system is controlled by a pilot model.
The authors declare that they have no known competing finan- We would like to thank Mr. Tsuneharu Uemura for the technical
cial interests or personal relationships that could have appeared to assistance and useful comments in the flight simulation with his
influence the work reported in this paper. experience being a skilled pilot.
R. Takase et al. / Aerospace Science and Technology 106 (2020) 106125 7
Fig. 10. States of FTFC systems (Fault No. 2: aileron & rudder effectiveness).
[6] M. Sato, G. Hardier, G. Ferreres, C. Edwards, H. Alwi, L. Chen, A. Marcos, Fault fication of learning systems in aerospace applications, in: Infotech@ Aerospace,
tolerance-background and recent trends, J. Soc. Instrum. Control Eng. 57 (4) 2005, p. 6912.
(2018) 279–286. [15] S. Bhattacharyya, D. Cofer, D.J. Musliner, J. Mueller, E. Engstrom, Certification
[7] Y. Zhang, J. Jiang, Bibliographical review on reconfigurable fault-tolerant control Considerations for Adaptive Systems, 2015.
systems, Annu. Rev. Control (2008). [16] R. Takase, J.O. Entzinger, S. Suzuki, Interaction between a fault-tolerant flight
[8] A. Belkharraz, K. Sobel, Simple adaptive control for aircraft control surface fail- control system using simple adaptive control and pilot’s pitch control, in: Asian
ures, IEEE Trans. Aerosp. Electron. Syst. 43 (2) (2007) 600–611. Control Conference, 2017.
[9] I. Gregory, E. Xargay, C. Cao, N. Hovakimyan, Flight test of an L1 adaptive con- [17] I. Barkana, Simple adaptive control - a stable direct model reference adaptive
troller on the NASA AirSTAR flight test vehicle, in: AIAA Guidance, Navigation, control methodology - brief survey, in: IFAC, 2007.
and Control Conference, 2010, p. 8015. [18] P.A. Ioannou, K.S. Tsakalis, A robust direct adaptive controller, IEEE Trans. Au-
[10] T. Nishiyama, S. Suzuki, M. Sato, K. Masui, Simple adaptive control with PID for tom. Control 31 (11) (1986) 1033–1043.
MIMO fault tolerant flight control design, in: AIAA SciTech Forum, 2016. [19] I. Barkana, Gain conditions and convergence of simple adaptive control, Int. J.
[11] M. Gao, J. Yao, Finite-time H ∞ adaptive attitude fault-tolerant control for reen- Adapt. Control Signal Process. 19 (1) (2005) 13–40.
try vehicle involving control delay, Aerosp. Sci. Technol. 79 (2018) 246–254. [20] I. Barkana, Robustness and perfect tracking in simple adaptive control, Int. J.
[12] Y. Xi, Y. Meng, Adaptive actuator failure compensation control for hypersonic Adapt. Control Signal Process. 30 (8–10) (2016) 1118–1151.
vehicle with full state constraints, Aerosp. Sci. Technol. 85 (2019) 464–473. [21] Y. Cao, J. Lam, Y. Sun, Static output feedback stabilization: an ILMI approach,
[13] H. Matsuki, T. Nishiyama, Y. Omori, S. Suzuki, K. Masui, M. Sato, Flight test Automatica 34 (12) (1998) 1641–1645.
of fault-tolerant flight control system using simple adaptive control with PID [22] D.T. McRuer, H.R. Jex, A review of quasi-linear pilot models, IEEE Trans. Hum.
controller, Aircr. Eng. Aerosp. Technol. 90 (1) (2018) 210–218. Factors Electron. (3) (1967) 231–249.
[14] S. Jacklin, J. Schumann, P. Gupta, M. Richard, K. Guenther, F. Soares, Develop- [23] M. Sato, A. Satoh, Flight control experiment of multipurpose-aviation-
ment of advanced verification and validation procedures and tools for the certi- laboratory-alpha in-flight simulator, J. Guid. Control Dyn. 34 (2011) 1081–1096.