1.
INTRODUCTION
Reconf igu ra ble
FI
ight
Control
Via
Multiple Model Adaptive
Control Methods
PEI’ER
S.
MAYBECK,
Fellow, IEEE
RICHARD
D.
STEVENS
Air
Force Institute
of
’Lechnology
A
multipk model adaptive controller (MMAC)
is
shown
to
provide
effective reconllgurabllity
whcn
subjected
to
single
ad
doubk
failures
of
8c11sors
and/or
actuators.
Its
performance
is
enhanced
by
M
alternate computation
of
the
MMAC
hypothesis
probabllitks,
use
of
maximum
a posteriori probability (MAP)
versus Bayesian
form of
the
MMAC (or a
modified
combion(ion
of
both), ad reduction
of
identillcation ambiguitks
througb
scalar
rcsMual
monitoring for
th
case
of
seosor failures.
Manuscript
received
April
1,1990;
revised
July
3,1990.
IEEE
Log
No.
44359.
Authors’ address Dept.
of
Electrical and Computer Engineering,
Air
Force Institute
of
’LechnologyENG, Wright-Patterson
Air
Force
Base,
Dayton,
OH
45433-6583.
0018-9251/91/0500-0470
31.00
@
1991
IEEE
~
470
IEEE TRANSACTIONS ON AEROSPACE AND ELEC’IRONIC
SYSTEMS
VOL.
27.
NO.
3
MAY
1991
For many applications, it
is
highly
desirable
to develop an aircraft flight control system with
reconfigurable capabilities: able to detect and isolate
failures
of
sensors and/or actuators and then to employ
a controller algorithm that has been specifically
designed for the current failure mode status. One
means
of
accomplishing this, in
a
manner that
is
ideally suited
to
distributed computation,
is
multiple
model adaptive estimation
(MMAE)
[1-4]
and control
@MAC)
i5-7.
Assume that the aircraft system
is
adequately
represented by a linear perturbation stochastic state
model, with an uncertain (failure status) parameter
vector affecting the matrices defining the structure
of
the model or depicting the statistics
of
the noises
entering it. Further assume that the parameters
can take on
only
discrete values; either this
is
reasonable physically (as for many failure detection
formulations), or representative discrete values are
chosen throughout the continuous range
of
possible
values. Then a Kalman filter
is
designed for each
choice
of
parameter value, resulting in a bank
of
K
separate “elemental” filters. Based upon the observed
characteristics
of
the residuals in these
K
filters, the
conditional probabilities
of
each discrete parameter
value being “correct”, given the measurement
history to that time, are evaluated iteratively.
A
separate set
of
controller gains is associated with each
elemental filter in the bank. The control value
of
each
elemental controller is weighted by its corresponding
probability, and the adaptive control
is
produced as
the probability-weighted average
of
the elemental
controller outputs.
As
one alternatiw (using
MAP
rather than minimum mean square root, or
MMSE,
criteria for optimality), the control value from the
single elemental controller associated with the highest
conditional probability can be selected as the output
of
the adaptive controller.
Recent research has applied an
MMAC
algorithm
with four elemental controllers to sensor or actuator
failure detection and control reconfiguration in a
STOL
F-15
aircraft
[8,
91.
Each elemental controller
was designed for a healthy aircraft, failed pitch rate
sensor, failed stabilator, or failed “pseudo-surface” (a
combination
of
canards, ailerons, and trailing edge
flaps).
This
initial feasibility study showed that the
elemental filters must be carefully tuned
to
avoid
masking
of
“good” versus “bad” models, and
this
specifically argues against loop transmission recovery
(LTR)
tuning. The published results
[S,
91
also provide
useful background on multiple model algorithms
and this specific application. The purpose
of
the
current research
[lo]
is
to improve the behavior
of
the
multiple model algorithm, to extend its applicability
to a wider range
of
failure conditions (including
soft
as well as hard failures, and multiple failures), and to
剩余10页未读，继续阅读
资源评论
WeiShi-TC2014-06-07一篇文章，分要得有点多
