Autonomous Precision Control APC on the OCP

1
Autonomous Precision Control (APC) on the OCP
OCP Meeting December 13-14, 1999 Dr. Mark
Campbell and Dr. Juris Vagners University of
Washington
2
Overview

• Technology Development Program Capabilities
• On-line Modeling
• Uncertainty Modeling
• Fault Detection and Control Reconfiguration
• Distributed On-line Modeling
• Current Control Design and Test Environment
• OCP Capabilities Desired
• Schedule

3
Program Objectives

• Develop an on-line system for autonomous high
  precision vehicles capable of
• monitoring and bounding system parameters
• creating a set of models
• adapting models to small parameter variations
• adapting models to system failures
• interfacing with precision controllers
• Extend the capability to multi-platform systems
  integrate with higher level management algorithms

4
Autonomous Precision Control at the Platform
Level
Fault Detection
Model Synthesis
Monitoring and Signal Processing
3
Control Redesign
Trajectory Optimization
Model Update
4
2
System
Controller
Trajectory
1

• Focus of the program is on blocks 2 and 3
• monitoring and uncertainty modeling
• model synthesis and update
• integrating with control redesign, fault detection

5
Capability 1 On-line Modeling

• Model Synthesis generates model from
  input/output data
• Model Update adjusts model to both small and
  large changes that occur at each time step
• Requires all sensor and command signals
• Handoff To Uncertainty Modeling, Fault
  Detection, and Control Reconfiguration

monitoring sensors
uncertainty modeling
Model Synthesis
Monitoring and Signal Processing
fault detection
feedback sensors
higher level management
Model Update
command/control signals
control reconfiguration
6
Current Work On-line Estimation of Nonlinearities

• Monitoring and bounding of uncertainties
• time-varying
• jumps
• mode changes

7
Current Work General Command Inputs
Slew command and process noise
Sensor signal and sensor noise

• Capable of using very general inputs and closed
  loop identification.

8
Capability 2 Uncertainty Modeling

• Develop a measure of uncertainty (errors/bounds)
  in system parameters within
• software (on-line modeling, control, and fault
  detection), sensors, external stimuli, etc.
• Requires all sensor and command signals, models
• Handoff To Fault Detection, and Control
  Reconfiguration, Higher level management.

monitoring sensors
fault detection
Uncertainty Modeling
feedback sensors
control reconfiguration
on-line models
higher level management
command/control signals
9
Concept for the On-line Uncertainty Model
Constant bounds based on actual noise/estimation
limitations
quick estimation of nonlinearity
convergence of bounds based on additional data
10
Capability 3 Fault Detection and Control
Reconfiguration

• Develop simple algorithms within APC framework
  which integrate on-line modeling (1) and
  uncertainty modeling (2) with
• fault detection
• control reconfiguration
• Integrate modeling capabilities (1, 2) with
  capabilities of SEC partners
• Requirements OCP is as modular as possible,
  interfaces are defined.

11
Capability 4 Distributed On-line Modeling

• Model Synthesis and Update for multi-ship cases.
• Uncertainty Modeling for multi-ship case
• Architecture of distribution to
• minimize communication
• distribute programming
• Integrate with higher level management
• Requirements TBD

12
Current Control Design and Test Environment

• MATLAB/Simulink environment (time varying UAV
  flight control simulation)
• Experimental tests are planned for a flexible
  two-link robotic arm and satellite - established
  testbeds that runs ControlShell on VxWorks,
  SystemBuild (MatrixX).
• A 3-DOF UCAV testbed will be constructed in Year
  3, run in a dedicated 3'x3' wind tunnel. The
  on-line software will be in ControlShell or
  dSPACE/Real Time Workshop/MATLAB.

13
Candidate OCP Capabilities

• A model of a UAV/UCAV consistent with the system
  to be built, especially actuators and sensors,
  and hardware characteristics such as sensor
  noise, etc.
• Could be released in different versions, each
  building in complexity. A simple Simulink (or
  other) model of what the OCP might look like
  would be great if released soon (spring 00?).
• A full simulation model of the hardware would be
  good by Oct 00.

14
Candidate OCP Capabilities

• The ability of software modularity. We know this
  is very difficult, but a modular system where ID,
  Fault Detection, and Control Adaptation are all
  from different teams would be great.
• Requires the developers of the OCP to define the
  integration between the components, such as an
  interface control document (ICD). These
  interfaces would need to be defined by Jan 01.

15
Candidate OCP Capabilities

• The ability to plug and play. If interfaces are
  defined, one could even envision the submission
  of MATLAB/Simulink software, which could then be
  ported to C using Real Time Workshop.
• We have done this in our lab with nice success
  (and the C code is really not too bad in terms of
  efficiency).

16
Schedule for APC
Validate
Integrate and
Validate on-line
monitoring system
validate fault
modeling (1)
in closed loop
detection software (3)
uncertainty software (2)
SEP 02
SEP 99
SEP 00
SEP 01
START
Simulate multi-ship
Integrate control
modeling
customization
uncertainty software (4)
software (3)