Project Title:
Neural Nets & Adaptive Reconfigurable Spacecraft Guidance and
Scientific Systems Company, Inc.
500 West Cummings Park
Suite 3950
Woburn, MA 01801
93-1-09 03 5355
Neural Nets & Adaptive Reconfigurable Spacecraft Guidance and
Control under Failure Conditions
Abstract:
Future NASA space missions call for unprecedented levels of
precision and reliability. Space Structures like other
engineering systems, will suffer from unexpected failures and
environmental disturbances. Although a large number of control
subsystems are used to nullify or ameliorate the effects of such
events, a health monitoring and supervisory control system is
required to constantly monitor the operating status of each of
these subsystems. Whenever a failure is predicted, the supervisor
will take corrective actions. A new technique is proposed here
for health monitoring and supervisory control which combines
techniques from the areas of Artificial Neural Nets (ANN) and
Robust Stochastic Control Theory. The proposed Phase I effort
will consist of the following tasks: (i) Identify and develop an
architecture for the health monitoring and supervisory control
system, (ii) Identify and develop techniques for detection and
classification using ANN after preprocessing by Extended Kalman
Filtering and Principal Component Analysis, (iii) Identify and
develop techniques for adaptive reconfiguration of control under
failure conditions, and (iv) Demonstrate the detection,
classification and reconfiguration for a sensor/actuator failure
on a NASA Test Article case study simulation. During Phase II,
other types of failures will be considered and the software will
be implemented on line and tested on a NASA Test Article.
The concepts of health monitoring and supervisory control can be
applied to solve wide range of problems such as those encountered
in factories, electric utilities and other large facilities.
Factories, for example, utilize many processes consisting of men
and machines, which can be decomposed in various layers.
Supervisory control can be exercised on each layer for efficient
operation. Similarly, in an electric power system, health
monitoring and supervisory control is needed for the reliable
operation of the massively large number of interconnected
subsystems.
Supervisory Control, Health Monitoring, Artificial Neural
Network, Stochastic Realization Theory