Design of control laws for automatic landing
(REAL)
DLR contributions to the REAL
project
Since the late sixties more and more aircraft types are equipped with
an autoland (automatic landing) system. Such a system allows an aircraft
to land safely even under extremely poor visibility conditions,
considerably increasing the aircraft's operability.
Autoland systems have to fulfil very high safety standards. These
standards are described in the so-called Joint Aviation Regulations - All
Weather Operations (JAR-AWO). The design of control laws for automatic
landing is an elaborate task, since many varying parameters, such as
aircraft loading, terrain and runway profiles, atmospheric conditions,
winds and turbulence, noise on guidance signals etc., have to be taken
into account (see the Figure below). In addition, the design models,
especially early in the aircraft development program, may show
considerable uncertainty. Usually, lots of manual design cycles and
considerable trial and error are required to finally meet the
certification criteria and to satisfy the monitoring pilots. Design model
updates may require additional design cycles.
Autoland Design, the REAL Project
The objective of the project REAL (Robust and Efficient Autopilot
control Laws design), funded by the Commission of the European Union, was
to investigate how modern robust control design methods can be used to
improve the efficiency of the autopilot control laws design process. As a
most challenging benchmark application, control laws for automatic landing
(autoland) were chosen.
The REAL-consortium consisted of DLR (Institutes of Robotics and
Mechatronics and Flight Systems , Airbus (France, Germany), NLR and
TU-Delft (The Netherlands) and ONERA (France). During the project,
autoland controllers for two dissimilar aircraft were developed, compare
the next Figure. The first, based on RealCAM, a civil aircraft model
provided by Airbus, allowed design teams from ONERA and DLR to apply and
refine their proposed approaches, especially to autoland-specific
requirements and issues. The second aircraft involved DLR’s ATTAS
(Advanced Technologies Testing Aircraft System), for which both design
teams had to deliver autoland control laws in a very short period of time
to prove the efficiency of their processes. The designs for ATTAS were
then tested in flight.
The REAL project
|
The design process had to fulfil the following criteria:
- systematic incorporation of design requirements,
- explicit consideration of uncertainties and parameter variations
from the start,
- the resulting control laws have to be visible and physically
understandable,
- a high level of automation of the design activities.
The benchmark problems involved the design of control laws for the
final approach (starting at ~1000 ft above the runway threshold) using the
Instrument Landing System and for the flare manoeuvre until touchdown of
the main wheels on the runway. Much uncertainty regarding aerodynamic
model parameters, strong wind and turbulence, as well as environmental
effects such as runway slope or irregular terrain in front of the runway
(as illustrated in the first Figure) had to be taken into account.
The DLR design approach
DLR enhanced and applied its flight control law design process, as
described here.
The proposed DLR flight control laws design process is based on the
long-standing experience of the Control Design Engineering group in the
field of object-oriented modelling and multi-objective optimisation. The
overall process is depicted in the following Figure. The principal steps
(left hand side), as applied to the autoland designs, will be explained in
the following subsections. To support the process, software tools and
methods are available or have been developed. These are depicted to the
right.
The DLR flight control laws design
process, supported by method and tool development
|
A/C Modelling
The aircraft models for the two benchmark problems, including
environmental effects, disturbances, etc. were implemented in Dymola using
the in-house developed Flight Dynamics Library, based on the
object-oriented modelling language Modelica , see below. The code for
simulation as well as the code for trimming (i.e. computing equilibrium
control deflections for a specific aircraft state) were generated from the
Modelica models and distributed with the benchmark problem descriptions
and software to all REAL project partners.
Controller structure synthesis
The first actual design step is the definition of the controller
structure. The structure as adopted for the autoland benchmarks is
depicted in the Figure below. Its functions have been grouped into three
main loops: stability and command augmentation, path / speed tracking, and
guidance (in the figure separated by colour backgrounds red, yellow, and
green respectively).
DLR autoland controller
architecture |
The task of the inner loop is to improve the stability and tracking of
attitude command variables. This part of the controller was designed with
Nonlinear Dynamic Inversion (NDI), which includes inverted model equations
in the control laws. For the two designs (RealCAM and RealATTAS), these
inverted equations were fully automatically generated from the aircraft
models in Modelica, avoiding any need for manual coding work.
The task of the path tracking loops is to make the aircraft follow
flight path and speed references. Four modes were designed: for the
approach phase the Total Energy Control System (TECS) was used for
decoupled tracking of flight path angle and speed commands, and a
classical PD control law was used for lateral flight path tracking.
Shortly before landing, the flare law, based on the so-called “variable
Tau” principle, takes over in order to reduce vertical speed to an
acceptable level for touchdown. The thrust is reduced simultaneously using
a retard function. Laterally, a classical align mode takes over from the
lateral path tracking mode in order to align the aircraft with the runway
centre line in case of cross wind, while keeping lateral deviation to a
minimum.
The task of the guidance loop is to derive flight path references from
guidance signals for the path tracking loops. For autoland these are
localizer (LOC) and glide slope (GS) radio signals. In order to improve
the estimation of metric deviations from the approach path, an altitude
over threshold estimation was implemented.
In the Feedback Signal Synthesis block air data measurements are
filtered complementary with inertial counterparts in order to reduce the
noise level due to turbulence. Also the side-slip angle is estimated for
use in the inner loops.
Multi-objective parameter
optimisation
In the proposed design process, free parameters in the control laws are
automatically tuned using multi-objective optimisation (MOPS ). In order
to handle complex control laws consisting of multiple interacting
functions, as is the case in autoland, a new tuning strategy was developed
that starts with tuning a single function, but sequentially leads to the
optimisation of all controller functions simultaneously. In the case of
autoland, for each controller function (see Figure above) an optimisation
sub-task was defined. This involved modelling of the detailed function
architecture and selection of appropriate design criteria for tuning and
compromising. Tuning then started with the inner loop functions. After a
satisfactory result had been achieved, the next function (i.e.
longitudinal or lateral path tracking) was added and another optimisation
was started. The optimiser may still adjust inner loop parameters, but by
retaining the inner loop design criteria as well, it is prevented from
distorting inner loop performance in case other outer loop functions (i.e.
flare or align) are connected. In the following steps glide slope and
flare/align modes were added, eventually leading to simultaneous
optimisation of all control law functions.
In the optimisation of glide slope and flare functions, nominal
performance was addressed via criteria from a single landing simulation.
In the proposed design process, each type of uncertainty and varying
parameters may be handled in the most natural way. Known varying
parameters are automatically addressed in the Dynamic Inversion-based
control laws. Uncertainty of model parameters is addressed by multi-case
optimisation, i.e. by compromise tuning for nominal and worst-case
parameter combinations simultaneously.Unspecified uncertainty (e.g.
unmodelled dynamics, time delays) is addressed by imposing sufficient
stability margins via design criteria. As a new contribution, unknown
varying disturbance parameters were addressed as in prescribed assessment
of autoland control laws, namely by incorporating risk criteria computed
from on-line Monte-Carlo analysis in the optimisation. In this way an
acceptable solution was found automatically, whereas otherwise fulfilling
risk requirements turned out to be difficult to achieve.
Controller Assessment
In the design process, the purpose of assessment is to detect hidden
weaknesses in the designed flight control system. If performance or
robustness is unsatisfactory at some point, optimisation criteria,
considered model cases or even the controller structure may have to be
adapted. In the REAL designs, the following means for assessment were
used:
- Parameter studies: A grid- or optimisation- based search via the
model parameters was performed to find cases where the controller
performance is at its worst. Worst-case models may serve to update the
set of cases used for multi-case robust design in MOPS.
- Monte-Carlo analysis: This is performed even more extensively than
in the optimisation. For example, limit risks are also assessed, where
one of the randomly varying model parameters is deliberately set to its
maximum or minimum value. This analysis is also prescribed by the
certification authorities.
- Desktop flight simulation and visualisation [link to modelling.doc]:
This tool was used to visualise simulation or even flight test data in
order to improve the understanding of control law behaviour, especially
under extreme conditions such as heavy cross wind or severe turbulence.
Real-time simulation was used to qualitatively test the inner loops by
interactively commanding attitude variables using a joystick. For the
REAL project, the Aviator’s Design Simulator (AVDS ) was used, see
below.
Results
After the RealCAM design was finished, the controller structure was
tuned for ATTAS using the developed optimisation set-ups in only a few
weeks, demonstrating a very short design cycle for a dissimilar aircraft.
The ATTAS design was successfully flight tested during six automatic
landings, see the next Figure, and the movie on the ATTAS page .
Both the RealCAM and ATTAS designs fulfilled practically all
performance and robustness requirements, based on JAR-AWO specifications.
Industrial evaluators found that the proposed design process fulfilled all
requirements listed at the top of this page.