Control of LargeScale Reconfigurable Systems

1
Control of Large-Scale Reconfigurable Systems

• Lara Crawford
• Lee Ackerson, Dave Biegelsen, Minh Do, Dave Duff,
  Craig Eldershaw, Markus Fromherz, Haitham Hindi,
  Johan de Kleer, Lukas Kuhn, Dan Larner, Hai
  Nguyen, Bryan Preas, Wheeler Ruml, Greg Schmitz,
  Lars Swartz, Armin Völkel, Rong Zhou
• Palo Alto Research Center

1
2
modularity communication coordination
2
3
Modular Reconfigurable Machines

• What if you could build machines like Lego
  structures?

• Integrated yet highly reconfigurable machines
• self-contained modules
• highly reusable components
• many configurations

4
Benefits of Modularity
Customizability
Robustness
Repairability
Reconfigurability
4
5
Benefits of Modularity
Customizability
Robustness
Repairability
Reconfigurability
5
6
Benefits of Modularity
Customizability
Robustness
Repairability
Reconfigurability
6
7
Benefits of Modularity
Customizability
Robustness
Repairability
Reconfigurability
7
8
Benefits of Modularity
Customizability
Robustness
Repairability
Reconfigurability
8
9
Challenges of Modularity
System must handle arbitrary configuration
Many ways of achieving a task ? difficult planning
?
Independently controlled units must be coordinated
System is subject to communication restrictions
Many independent units ? difficult to isolate
failures
System must be scalable
System must respond rapidly to exceptions
10
Outline

• Highly reconfigurable printing machines
• Design principles control architecture
• Planning scheduling
• Control coordination
• Diagnosis
• Conclusions

11
High-End Printing Systems

• A complex multi-function device
• variety of operations
• complex interactions
• real-time, high-speed operations
• Control software ensures performance and
  correctness

Xerox iGen
11
12
Toward Highly Reconfigurable Machines

• Hypermodular paper path
• small, self-contained modules
• highly reusable components
• few module types, many configurations

Conceptual not a product!
A revolutionary concept integrated, yet highly
reconfigurable
Requires on-line planning and tightly coordinated
control
13
Hypermodular System at PARC
Not a product!
distributed actuation sensing processing
14
Hypermodular System at PARC
(photo removed)
A reconfigurable high-speed printing system
14
15
Design Principles

• Enabling reconfigurability
• Software and hardware modularity
• Model-based approach
• Multi-scale control, distributed control
  architecture, control coordination
• Encapsulation of knowledge
• Explicit contracts
• Delegation of responsibility
• Autonomy
• Escalation
• Technology designed for printing domain but is
  generally applicable to other domains

16
Control System Architecture
job description
planning and scheduling
coordination
actuator control
sheet motion
17
Planning Scheduling for Printers

• Quickly find online
• Legal sheet itineraries
• Maximize throughput

• Given
• Machine model (at startup)

• Sheet desired (online)
• Machine state (online)

18
Planning Challenges

• Compositional modeling
• describing machines, actions, and constraints
  from the ground up
• Real-time planning
• very fast real-time planning for many routes and
  machine states
• On-line planning
• simultaneously handling new jobs, new plans, and
  parallel execution
• Exception handling
• recovering from faults on the fly without
  operator involvement

19
Modeling Actions and Jobs
Sample job description
Sheet-23 initial
Location(Source) goal
Location(Destination) FacingUp(Front)
HasImage(Front, Im-17)
no function in structure
20
Planning for Real-Time Execution
Feeder
Finisher
Regression planner using A for tree search with
admissible heuristic on plan end time (wall-clock)
21
Simulation Hypermodular System
(movie removed)
22
In-Flight Sheet Rerouting
2
Finisher 1
1
Finisher 2
Purge tray
3

• Blue job integrity is violated
• Sheet 3 (green) jams

23
Replanning Paper Jam
(movie removed)
24
Replanning In-Flight Rerouting
(movie removed)
25
Control System Architecture
job description
planning and scheduling
coordination
actuator control
sheet motion
26
System Features
Highly modular paper path

• Local actuation via nips (rollers)
• Local sensing via sheet edge detection
  (asynchronous measurements)
• Local controller processing
• Networked communication
• Highly distributed, large scale (hundreds of
  actuators sensors)

High-speed operation, tight cooperation
26
27
Control Challenges

• Networked communication
• delays and bandwidth limitations
• Compositional modeling knowledge
• supporting a wealth of configurations and module
  types
• Exception handling
• recovering from faults on the fly without
  operator involvement

• Distributed coordination
• dynamic, high-speed, tight cooperation among
  modules

28
Coordinating Control Architecture

• Coordination task
• up proxy for sheet control
• down module coordinator
• Sheet controller roles
• interpret, distribute plans
• share feedback
• monitor sheet progress

Planner
Encapsulation, delegation, autonomy, escalation,
coordination
29
Interfaces
Planner
29
30
Networked Controller Coordination

• Sheets in multiple nips simultaneously
  
• dynamic control group

control group synchronization
initialization
maintenance
30
31
Physical-Level Control

• Problem
• Given a set of way points (t, x), find motor
  commands to track these waypoints
• Subject to actuator limitations
• With incomplete state information and infrequent
  sensor data
• In the presence of disturbances
• Solution
• Time-optimal servo control
• Minimizes time required to reduce error to zero
• Optimal control based on model of dynamics
• Actuator limits handled explicitly

32
Integrated Hypermodular System
(movie removed)
33
Control System Architecture
job description
planning and scheduling
coordination
diagnosis
actuator control
sheet motion
34
Diagnosis

• Goal
• Quick accurate repairs
• Robustness to failed components

• Challenges
• Symptoms often caused by non-local
  intermittent faults
• Responding to surface symptoms can cause more
  problems
• Solution
• Diagnosis needed
• Diagnosis Testing Reasoning
• Selecting tests to optimize information gain
• Inferring hidden faults from test results

35
Diagnosing Highly Modular Systems
36
Research Issues

• Architecture design and analysis
• Modeling and model abstraction
• Simulation and verification

37
Toward Highly Reconfigurable Systems
A revolution inmodularity and
control.Changing the wayproducts are put
together.
38

• Thank you!