INTELLIGENT SYSTEMS I

1
INTELLIGENT SYSTEMS - I
Workshop on European Space Information
Technology in the 21st Century 27-28th September
2000

• Roger Thompson and Roger Ward
• Science Systems (Space) Ltd.
• Methuen Park
• Chippenham, Wiltshire, UK SN14 0GB
• 44 (0)1249 466466
• www.scisys.co.uk

2
Introduction

• Definition of Intelligent Systems
• On-board Intelligence
• Ground-based Intelligence
• Importance of Space-Ground Integration
• Appropriate Applications
• Appropriate Technologies
• Validation of Intelligent Systems
• History Some Examples
• The Way Forward

3
The Vision...

• Scientists controlling their own telescopes
• Spacecraft seeking targets of opportunity as they
  arise
• Interoperable space and ground segment automation
• Satellites that call home when they have a
  problem, or call the end user when they find an
  anomaly
• Formations of spacecraft on "patrol" passing
  guidance and actions from one to another
• Multiple satellites operating as conventional
  networks, "sharing the processing load
• Long duration missions augmented with latest
  technology
• Reduced Operations Costs - Lights Out in the
  Control Room

4
The Reality...

• The impossible is becoming increasingly possible
• Software is becoming increasingly Intelligent
  with the application of advanced software
  technology
• On-board software is the only subsystem which can
  be augmented post launch
• Software can be used to maximise scientific
  return, and make in-orbit assets more accessible
  to investigators
• But
• Software must be actively considered in Mission
  Design
• European industry is Good at Software - but the
  Initiative is being taken elsewhere...

5
What is an Intelligent System?

• Key Features
• Reasoning or Decision Making without Human
  Intervention
• Delegation of Operational Authority from Human
  Operators to the System
• Abstraction of Operational Requests (High-Level
  Goals)
• Autonomous Response to Observable Events
• Ability to Handle Failures/Anomalies
• Intelligent or Autonomous? A matter of degree
• Location - can be anywhere in the System
• On-board (OBS) - Autonomy
• Ground Control - Automation
• Technology
• Invariably Software
• Can be Traditional (C, ADA)
• AI Technologies may allow more efficient
  Development

6
On-board Intelligence (Autonomy)

• Make Impossible Possible
• Rapid Reaction to Events - Propagation Delays and
  Non-Contact Periods
• Simplify Operations
• Goal Directed Commanding, Target Selection
• Autonomous Scheduling/Re-scheduling of Operations
  based on Events
• On-board Quality Control / Data Reduction
• Qualitative or Fuzzy FDIR cope with sensor
  degradation
• Benefits
• Enhanced Performance or Science Return
• Reduction of Data Traffic on Space-Ground Link
• Open Accessibility of Payload Operations -
  Telescience
• Enhanced Safety/Reliability
• Issues
• Complexity of Ground Operations
• Failure of Autonomy - Safety, Diagnosis/Correction
  , Fall-back Ops Mode

7
Ground-based Intelligence (Automation)

• Automation of Operations
• Optimisation and Generation of Mission Timelines
• Management of Operational Constraints (Time and
  Resource)
• Automated Operations Schedules
• Automated Operations Procedures
• Anomaly Detection and Automated Recovery
• Fault Diagnosis
• Benefits
• Fewer Operators or Lights Out Operations
• Increased Reliability of Routine Operations
• Potential for Improved Service Provision
• Reduction of through-life Operations Costs
• Issues
• Validation
• Failure of Automation - Safety, Ability of
  Engineers to React, Potential Service Loss
• Appropriate Displays and Operations History

8
Importance of Space-Ground Integration

• Cohesive System - not two independent halves as
  at present
• Migration of Functions from Ground to Space
• Common (or Equivalent) Infrastructure on-board
• Basic TM/TC processing On-board
• HCI for Operations more useful on Ground!
• Visibility of On-board Decisions enhanced if SCC
  OBS share common Representation
• Need to Standardise Higher Levels of Commanding
  and Reporting
• Command, Event, Parameter, Scheduled Items
• Common Model of Procedure/Function and Reporting
  against this
• Definition of APIs rather than Data Formats
• Support for Interoperable Software Components
• Remote Agents DS-1
• Technologies
• CORBA
• TCP/IP or SCPS

9
Potential Integrated Model of Operations
Parameters Commands Events
Ground Segment
10
Need to Identify Appropriate Applications

• Autonomous Operations
• Goal-oriented Tasking
• Reaction to Events (e.g. imaging of celestial
  events)
• On-board Scheduling (e.g. EO constraints - cloud
  cover, mode, power)
• Management of Resources Power, Fuel, Downlink
• Task Cueing within Constellation or between
  Instruments
• Fault Management FDIR, Fault Diagnosis
• Data Reduction
• On-board Processing (e.g. SAR Feature Extraction)
• On-board Quality Control (e.g. discard obscured
  images)
• Housekeeping Data
• Autonomous Flight Control
• Agile AOCS / Image Motion Compensation
• Formation Control
• Constellation / Formation Data Management
• Pass Management
• Unreliable Ground Contacts - Pass Rescheduling,
  Alternative Comms Routes
• In Theatre Data Delivery

11
Need to identify appropriate technologies

• Technologies for Ground Systems will be discussed
  shortly
• Possible Technologies for end-to-end integration
  
• CORBA
• SCPS or Internet protocols for distributed access
  
• Intelligence may be distributed on the ground
• Interaction and control of on-board autonomy
• Standardisation of Ground and Space Components
  through OMG or CCSDS?

12
Need to identify appropriate technologies

• Possible Technologies for On-Board Software
• Classical coding (Ada or C)
• Heuristic
• Fuzzy Logic
• Neural Networks
• Potential Functions
• Only Classical coding (with auto code) being
  exploited currently in Europe, US more
  adventurous?
• Need to adequately address these technologies

13
Heuristic Techniques

• Knowledge Based, Constraint Based or Rule Based
• Allows experience and knowledge to be
  incorporated
• Solve highly constrained, high data volume
  problems
• Applicable to Diagnostics, Decision making
  problems and Spacecraft Scheduling

14
Fuzzy Controllers

• Allows the control of complex systems
• Fuzzy logic provides an efficient approximate
  description of a system.
• Can program using algorithms then tune it based
  on input data and expert knowledge.
• Widely used on systems such as metros, ABS on
  cars
• TUD recently proposed for AOCS

15
Neural Networks
x0

• Non deterministic suitable for situations with
  many inputs possibly uncertain inputs
• Needs to be trained
• Testing is a major problem
• Commonly used for diagnostics
• Could be used to detect Solar flares, space
  debris etc.

f(x)B
f(x)A
x1
y
f(x)A
f(x)B
f(x)C
f(x)A
f(x)B
x2
1
1
16
Potential Functions

• Based on Lyapunovs Second Method
• Allows control of complex Non-Linear Systems
• Will solve large multi-variable systems through
  control of single potential function
• Useful for Formation Flying, RVD...

17
What would make them appropriate?

• They must be better than classical approaches
• Traditionally they should be
• Deterministic
• Efficient
• Secure
• Supported for radiation hard processors etc
• Testable
• Take a Systems View

18
How do we Test these things?

• Classic AI problem
• may not be deterministic
• even if solution is deterministic may be too many
  paths.
• how rigorously was the human operator tested
• test specific scenarios, rather than all possible
  conditions
• test using a subset of the data used for
  training.
• Do systems level criticality analysis
• level of testing should depend on level of
  criticality
• Build confidence slowly. If mission is not
  possible any other way maybe just accept it.

19
Examples of Autonomy

• XMM
• Real time spacecraft under ground control
• Survived at other times
• AOCS
• Autonomous Momentum Dumping, FDIR
• FIRST
• 48 hours nominal operation without ground contact
• Operations based on a Timeline
• Beagle 2
• Ground contact approximately 15mins every 4 days
  not real time
• Nominal Operation based on timelines
• Battery monitoring, Safe Mode

20
Ground/Space Integration on STRV - RATE

• Space Segment
• Attitude Determination Remote Agent (ADRA)
• Migrate existing Attitude Determination Algorithm
  from Ground to Space Segment
• Uses CORBA Integrated with SCPS
• Ground Segment
• STRV MCC
• Bridge to CORBA based UNiT
• ADRA becomes a component of the Ground Segment

21
Benefits

• Intelligent systems are increasingly used as part
  of Ground Segment
• Only classical software technologies used
  on-board
• Ground-Space Integration is not addressed

22
Conclusions

• Need mission developers to recognise
  opportunities for intelligent software technology
  and to support its provision.