Towards a State Based Control Architecture for Large ...Europe-US-Japan Started 1998, Early Science...
24
Presented at ICALEPCS, Oct 14th 2011 1 of 16 Robert Karban, Nicholas Kornweibel (ESO, Garching, Germany) Daniel Dvorak, Michel Ingham, David Wagner (Jet Propulsion Laboratory, California Institute of Technology, Pasadena, CA, USA) Towards a State Based Control Architecture for Large Telescopes: Laying a Foundation at the VLT
Transcript of Towards a State Based Control Architecture for Large ...Europe-US-Japan Started 1998, Early Science...
Presented at ICALEPCS, Oct 14th 2011 1 of 16
Robert Karban, Nicholas Kornweibel (ESO, Garching, Germany)Daniel Dvorak, Michel Ingham, David Wagner (Jet Propulsion Laboratory, California Institute of
Technology, Pasadena, CA, USA)
Towards a State Based Control Architecture for Large Telescopes: Laying a Foundation at the VLT
Presented at ICALEPCS, Oct 14th 2011 2 of 16
About Robert Karban
Presented at ICALEPCS, Oct 14th 2011 3 of 16
Outline
• Context
• Architecture
• State Analysis
• VLT Field Testing
• Summary and Future Work
Presented at ICALEPCS, Oct 14th 2011 4 of 16
Context - ESO major projects
Very Large Telescope (VLT)Started 1988, in operation since 1999
Atacama Large Millimeter Array (ALMA)
Europe-US-JapanStarted 1998, Early Science Operations started in October 2011.Completion expected in 2012Images on this slide were produced by ESO
Presented at ICALEPCS, Oct 14th 2011 5 of 16
Context - The E-ELT
Images on this slide were produced by ESO
• 40 m class mirror• Will be the largest
optical/near-infrared telescope in the world
• Gather 15 times more light than any other telescope today.
• Exciting science: extra solar planets and discs, galaxy formation, dark energy/dark matter, and frontiers of physics.
• If approved construction could start in 2012 with beginning of operations 2020-2022
Presented at ICALEPCS, Oct 14th 2011 6 of 16
VLT Wavefront control E-ELT Wavefront control
• 10000 tons of steel and glass• 20000 actuators, 1000 mirrors• 50000 I/O points, (M1 has 15000 alone)• Large data volume (700Gflops/s, 17Gbyte/s),
only engineering data• Multitude of interacting, distributed control
loops (0.01Hz->kHz rates)• Overall function and performance of the
telescope is allocated to the control system
Context - Challenges for the Control System
• 600 tons of steel and glass• 200 actuators, 3 mirrors• 2000 I/O points• Small data volume• Some interacting, distributed control
loops (0.01Hz->50Hz)• Overall function and performance of the
telescope is allocated to the control system
Presented at ICALEPCS, Oct 14th 2011 7 of 16
VLT Wavefront control E-ELT Wavefront control
• 10000 tons of steel and glass• 20000 actuators, 1000 mirrors• 50000 I/O points, (M1 has 15000 alone)• Large data volume (700Gflops/s, 17Gbyte/s),
only engineering data• Multitude of interacting, distributed control
loops (0.01Hz->kHz rates)• Overall function and performance of the
telescope is allocated to the control system
Context - Challenges for the Control System
• 600 tons of steel and glass• 200 actuators, 3 mirrors• 2000 I/O points• Small data volume• Some interacting, distributed control
loops (0.01Hz->50Hz)• Overall function and performance of the
telescope is allocated to the control system
Presented at ICALEPCS, Oct 14th 2011 8 of 168
Architecture Investment “Sweet Spot”
0%
20%
40%
60%
80%
100%
120%
0% 10% 20% 30% 40% 50% 60% 70%Fraction budget spent on architecture
(Equations from Reinholtz ArchSweetSpotV1.nb)
Frac
tion
bu
dget
spe
nt
on r
ewor
k+ar
chit
ectu
r
KSLOC 10 KSLOC 100 KSLOC 1000 KSLOC 10000
10K SLOC
100K SLOC
1M SLOC
10M SLOC
Source: Kirk Reinholtz, JPL
Example:For 1M lines of code, spend ~29% of s/w budget on architecture for optimal ROI
Predictions from COCOMO II model for software cost estimation
Fraction of budget spent on architecture
Frac
tion
of b
udge
t spe
nt o
n re
wor
k +
arch
itect
ure
Presented at ICALEPCS, Oct 14th 2011 9 of 169
Architecture Investment “Sweet Spot”
0%
20%
40%
60%
80%
100%
120%
0% 10% 20% 30% 40% 50% 60% 70%Fraction budget spent on architecture
(Equations from Reinholtz ArchSweetSpotV1.nb)
Frac
tion
bu
dget
spe
nt
on r
ewor
k+ar
chit
ectu
r
KSLOC 10 KSLOC 100 KSLOC 1000 KSLOC 10000
10K SLOC
100K SLOC
1M SLOC
10M SLOC
Source: Kirk Reinholtz, JPL
Trend:The bigger the software, the bigger the fraction to spend on architecture
Example:For 1M lines of code, spend ~29% of s/w budget on architecture for optimal ROI
Predictions from COCOMO II model for software cost estimation
Fraction of budget spent on architecture
Frac
tion
of b
udge
t spe
nt o
n re
wor
k +
arch
itect
ure
Inference:Prior investment in a reference architecture pays dividends
Presented at ICALEPCS, Oct 14th 2011 10 of 16
Architecture - Goals
• Address the functional need derived from the wave front control strategy
• Contain system complexity• Promote modifiability and
scalability (and long-term maintainability)
• Enable high availability and fault tolerance
• Conceptual Integrity
Presented at ICALEPCS, Oct 14th 2011 11 of 16
Conceptual Architecture• Data driven• Decentralized• Separate Domain
Knowledge
• Integrate heterogeneous Control Systems
• Define a framework and design rules
Presented at ICALEPCS, Oct 14th 2011 12 of 16
State Analysis – What does it mean?• Imagine driving a car.
As a driver you:• Have destinations and deadlines• Plan a route• Rely on gauges and your own senses
• In other words, you:• Set objectives regarding the state of
the world• Monitor the state of the world• Form a coherent notion of the state of
the worldand you anticipate its changes
• State is central
Presented at ICALEPCS, Oct 14th 2011 13 of 1613
State Analysis - The “Control Diamond”
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 14 of 1614
State Analysis - The “Control Diamond”
Our model of how the world works
helps us make sense of our senses
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 15 of 1615
State Analysis - The “Control Diamond”
Our model of how the world works
helps us make sense of our senses
We see, not what is, but what we perceive
— with a little help from our senses
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 16 of 1616
State Analysis - The “Control Diamond”
We react, not to things as they are, but rather to things
as we perceive them
Our model of how the world works
helps us make sense of our senses
We see, not what is, but what we perceive
— with a little help from our senses
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 17 of 1617
State Analysis - The “Control Diamond”
We react, not to things as they are, but rather to things
as we perceive them
Our model of how the world works
helps us make sense of our senses
We see, not what is, but what we perceive
— with a little help from our senses
Our actions are guided by what we expect them to do,
given what we know
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 18 of 1618
State Analysis - The “Control Diamond”
We react, not to things as they are, but rather to things
as we perceive them
Our model of how the world works
helps us make sense of our senses
We see, not what is, but what we perceive
— with a little help from our senses
Our actions are guided by what we expect them to do,
given what we know
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 19 of 1619
State Analysis - The “Control Diamond”
We react, not to things as they are, but rather to things
as we perceive them
Our model of how the world works
helps us make sense of our senses
We see, not what is, but what we perceive
— with a little help from our senses
Our actions are guided by what we expect them to do,
given what we know
Real world
Similar behaviors
Virtual world
Presented at ICALEPCS, Oct 14th 2011 20 of 16
State Analysis – in a Nutshell
System UnderControl
StateControl
Hardware Adapter
Activity Planning & Execution
ControlGoals
Sense
StateEstimation
Act
Measurements& Commands
Commands
StateFunctions
StateValues
KnowledgeGoals
State Knowledge
Models
Clear delineation between control system and system under control
State is explicit
Separation of estimation from control
Models inform all aspects of control system
System operation via explicit, objective statements of intentControl System
Direct correspondence between physical state variables and software state variables
Presented at ICALEPCS, Oct 14th 2011 21 of 16
VLT Field Testing - Motivation
• Test E-ELT technology decisions
• Refurbish VLT control system
• Operational Environment
• Apply State Analysis Method
Presented at ICALEPCS, Oct 14th 2011 22 of 16
VLT Field Testing -Enclosure Control System Upgrade
• 1500 I/O points (Dome,
Windscreen, Louvers, etc)
• Interface to existing sensors and
actuators
• Driven by SA Control Diamond
• Estimators implemented with
LabView
Presented at ICALEPCS, Oct 14th 2011 23 of 16
VLT Field Testing - Main axes Control System Upgrade• Provides all means for telescope positioning
• Apply more rigorously SA, integrated with OOSEM
Presented at ICALEPCS, Oct 14th 2011 24 of 16
Summary and Future Work
• Summary
– SA is built on sound theory
– Guided by Architectural Principles
and Rules
– Confidence gained during VLT field
tests
• Future Work
– Collaboration between ESO and JPL
– SA profile for SysML
– Integration with MBSE practices
