FLEXCON

1
FLEXCON

• Flexible Embedded Control Systems

2
FLEXCON Real-Time Control
Real-Time Computing
Control Engineering
Control in Real-Time Computing
Real-Time Techniques in Control System
Implementation
3
Temporal Non-Determinism

• Decreases
• improvements in worst-case analysis methods
• tool development
• development of more deterministic implementation
  techniques
• Increases
• developments in general purpose computer systems
• new types of applications, e.g., Internet-based,
  operating in open and unpredictable environments
• next generation micro-chips
• stochastics will play a larger role
• sacrifice temporal determinism to maintain
  functional determinism

Increasing, at least for non-critical systems.....
4
Flexibilitet

• F m.a.p osäkerhet om resursutnyttjande
• F m.a.p osäkerhet om egenskaper hos
  implementationsplattform
• F m.a.p. osäkerhet om extern omgivning
• F m.a.p. osäkerhet om tasks (last)
• F m.a.p. specifikationer (interval/max/min vs
  fixa värden)
• F m.a.p. dynamisk systemuppdatering (plugn play)
  (komponenter, applikationer, systemprogramvara)
• F. i bemärkelsen event-triggered vs
  time-triggered (dynamic vs static)
• F. i utvecklingsprocessen (vid design-time),
  använda komponenter etc, konfigurering,
• F. m.a.p. virtuell resp fysisk miljö

5
WP1 Flexibility in real-time embedded control
system design using COTS platforms, languages and
components

• Component Technology (Ivica Crnkovic)
• embedded control systems
• real-time issues
• flexibility
• PhD student Johan Fredriksson (2003) (SAVE)
• Language Technology Java (Klas Nilsson)
• dynamic aspects
• flexibility
• PhD student Sven Gestegård Robertz
• Cont. of ARTES project
• Feedback scheduling in dynamic memory allocation
  (RT-Java)

6
WP23

• WP2 Control-Based Approaches in Embedded Systems
• WP3 Quality-of-Service and Resource Negotiation
  in Embedded Control
• Combined into a single WP with focus on control
  systems

7
Temporal Determinism

• Computer-based control theory is based on
• equidistant sampling
• negligible input-output latencies that can be
  ignored or constant latencies that easily can be
  compensated for
• Reality
• Varying execution times due to preemption,
  blocking, data-dependencies, caches, pipelines,
  network communication,
• Result
• Sampling interval jitter
• Non-negligible and varying latencies

8
Control Community
A new implementation and resource-aware control
paradigm is needed!
Resource-Constrained Control
9
Hard Control Implementation Approach

• Strive to maximize the temporal determinism
• E.g. using time-triggered and synchronous
  programming models
• Pros
• Simplifies attempts at formal verification for,
  e.g. safety-critical applications
• However, a large amount of hard real-time
  control applications are not safety-critical
• Cons
• Often requires special purpose solutions, i.e.,
  less efficient and more expensive
• Requires complete knowledge about resource
  utilization, load, ..
• May result in under-utilized systems with
  possibly poor control performance

10
Hard R-T Task Model

• Periodic/sporadic tasks with constant period,
  hard deadline, and known WCET
• Just a model
• Does not fit all control problems
• E.g. hybrid controllers, event-based controllers
• Overly restrictive for most control problems
• a missed deadline no catastrophy
• a late control signal is better than no signal at
  all

11
Soft Control Implementation Approach

• View the temporal nondeterminism caused by the
  implementation platform as an uncertainty or
  disturbance acting on the control loop
• Use control-based approach
• Inherent robustness of feedback
• Design for robustness against implementation
  uncertainties
• Active compensation, cp feedforward from
  measurable disturbances and adaptive control

12
Implementation-Robust Control
A tremendous amount of theory for plant
uncertainties
?
?
Very little theory for implementation platform
uncertainties
13
Implementation-Robust Control

• Temporal robustness
• timing variations
• Theory that allows us to decide which level of
  temporal determinims that a given control loop
  really requires in order to meet given objectives
  on stability and performance
• Is it necessary to use a time-triggered approach
  or will an event-triggered approach do?
• How large jitter in sampling interval and i-o
  latency can be tolerated?
• Is it Ok to now and then skip a sample?
• ..
• Functional robustness
• Fault-tolerance towards computer-level faults
  leading to data errors
• An increasing problem in future deep sub-micron
  technology hardware

14
Resource Allocation as a Control Problem

• In an applications with multiple (control) tasks
  the dynamic allocation of resources to the tasks
  can be viewed as a control problem in itself!
• The control performance can be viewed as a
  quality-of-service attribute (Quality-of-Control)

15
Control in Real-Time Computing

• Use of control-based approaches for uncertainty
  management in large real-time computer and
  communication systems is receiving increased
  attention
• The worst-case approach no longer feasible
• Feedback, feedforward, ...
• Control-oriented models capturing dynamics

16
Feedback Scheduling

• Dynamic on-line allocation of computing resources
• Feedback from actual resource utilization
• In principle, any computing resource

17
Feedback Scheduling Structures

• Feedback
• Reactive
• Feedforward
• Proactive
• Mode changes and admission control

18
Requirements on Scheduling Theory

• Relax the standard hard-real time assumptions
• Theory that better matches the needs of control
  systems

19
Requirements on Control Theory

• Co-design methods
• control design methods that take resoure
  constraints into account
• Improved understanding of how temporal
  non-determinism effect control performance
• analysis methods
• Tools
• Theory for aperiodic systems

20
Examples of recent developments

• Jitterbug (Cervin, Lincoln)
• Matlab toolbox
• analysis of how sampling period and i/o delay
  distributions effect control performance
• TrueTime (Cervin, Henriksson)
• Simulink toolbox
• co-simulation of temporal effects of real-time
  kernels and communication networks, and control
  performance
• New simple stability results (Lincoln)
• control loops with variations in delay
• networked control loops

21
Jitterbug
22
TrueTime
23
Tool Usage
Simulation withTrueTime
Analysis withJitterbug
SchedulingParameters (T,D,Prio, )
Task TimingParameters (latencies, jitter, )
ControlPerformance (variance, rise time,
overshoot, .)
Non-trivialrelationship
Complex, nonlinearrelationship
24
WP4 Testing-Based Verification and Monitoring of
Embedded Control Systems

• Högskolan i Skövde (Sten Andler)
• Focus on event-driven control systems
• Run-time properties for testability
• Test case selection and generation.
• Connection to MdH (Thane)

25
WP5 Robotics and Automation Demonstrator

• Common platform for demonstrating project results
• Maintain the project focused
• Not a moon-lander demonstrator
• Based on Robotics Laboratory in Lund (Klas
  Nilsson)
• EU project HRTC, incl. TTP
• EU project AUTOFETT with ABB,
• Strong links to ABB
• EU project SMErobot starting...

26
Thing in common with Saab

• Separation of concerns, modularity
• Multi-CPU VMEPCI/PMC systems
• Dependable communication/control
• COTS hardware
• Long-lasting platforms
• Safe execution
• Testing and monitoring
• Combination with formal methods desired
• Engineering efficiency/practices....

27
Possible SaabTech Issues

• Flexibility techniques for improved robustness
• Combining for mission-critical systems
• Safe languages (Well-defined execution)
• Certified run-time techniques (VMHW)
• Safe partitioning with shared resources.
• Formal verification (FLEXCONSAVE)
• Improved testing techniques (pre-runtime)
• Embedded on-line monitoring (run-time)
• Questions?