The Development of Neural Network Based System Identification ...

Chapter 8
CONCLUSIONS AND FUTURE WORKS
In this thesis, the system identification and control algorithms based on NN methodology
were developed to control an unmanned helicopter system in hovering flight condition.
The proven ability of the NN based system identification and controller design approach
to adapt to the dynamic system changes allows for the development of a flexible and
self-tunable flight controller that can be deployed into different UAS airframes in shorter
development time. Furthermore, the fast and simpler development of the dynamic model
using the NN approach offers significant benefit which reduces the cost associated with
the development of large aerodynamic databases. The proposed system identification
and NNAPC flight controllers for autonomous hovering were implemented and tested in
simulation environment and indoor flight test facility.
A detailed overview of the unmanned helicopter platform as well as the avionics
hardware used was provided. To prevent fatal crashes or damages to the helicopter
system due to probable hardware failures or programming mistakes, a novel safety
test rig was developed to restrict the movement of the helicopter during testing. The
development of the safety test rig allows the helicopter system to move freely in 6 DOF
flight. The test rig also was equipped with necessary instrumentation that provides the
end users with position and attitude information of the helicopter.
Several NN based system identification algorithms were developed to model the
unmanned helicopter dynamics from the flight test data. Three types of NN architectures
were deployed to model the dynamics of the helicopter; namely the MLP, HMLP and
modified Elman networks. The nearly optimal or optimal model structures for the
proposed NN architectures can be found based on the proposed model validation tests.