Adaptive neuro-fuzzy control approach for spacecraft maneuvers
This paper introduces a new technique to control spacecraft maneuvers. The new technique is based upon using neuro-fuzzy approach to predict the required control torque, using a modelless-strategy, for attitude and rate tracking subjected to torque constraints. The Neuro-Fuzzy Controller (NFC)...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
ICGST-ACSE
2006
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/23319/ http://irep.iium.edu.my/23319/ http://irep.iium.edu.my/23319/1/ACSE-2006.pdf |
| Summary: | This paper introduces a new technique to control
spacecraft maneuvers. The new technique is based upon
using neuro-fuzzy approach to predict the required
control torque, using a modelless-strategy, for attitude
and rate tracking subjected to torque constraints. The
Neuro-Fuzzy Controller (NFC) is built up using the
Adaptive Neuro-Fuzzy Inference System (ANFIS) which
transforms a fuzzy controller into an adaptive network to
take the advantage of all the neural network control
techniques proposed in the literature. First, the inverse
dynamics of the spacecraft is developed by training the
ANFIS with specified states such as Euler angles and the
angular velocities. These data can be collected via direct
measurements, estimators, or simulation using attitude
propagators. Second, three types of controllers are
developed, started with a Single Level NFC (SLNFC) to
a Multi Level NFC (MLNFC) and ended by a Hybrid
Controller. The configuration of the first and second
controllers depends on the structure of the data used in
the training phase. While, the hybrid controller utilizes
the NFC in general to solve the problem of large angles
attitude tracking in the absence of the system model and
brings the system to a steady state with relatively small
errors then, it switches to either a classical or a modern
controller to refine the steady state errors. Finally, each
one of them is tested against two different controllers
belonging to classical and modern control approaches for
the purpose of performance evaluation. The first one is a
classical PD controller using quaternion feedback, and
the other is a Non-Linear Predictive controller (NLP)
which is developed to predict the required control action
to track a certain trajectory under rate and torque
constraints. The developed controllers have shown a
competitive performance to that of classical one and the
simulation results give neuro-fuzzy control approach an
edge over the modern control approaches specially when
considering the hard constraint of a modelless spacecraft. |
|---|