2004-11-23 @ SaabTech Jönköping FLEXCON Flexible Embedded Control Systems.

2004-11-23 @ SaabTech Jönköping

FLEXCON

Flexible Embedded Control Systems


FLEXCON = Real-Time & Control

Real-TimeComputing

ControlEngineeringFLEXCON

Control in Real-Time Computing

Real-Time Techniques in Control System Implementation


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.....


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 (plug’n 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


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)


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


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


Control Community

A new implementation and resource-aware control paradigm is needed!

Resource-Constrained Control


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


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


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


Implementation-Robust Control

A tremendous amount of theory for plant uncertainties

?

Very little theory for implementation platform uncertainties

?


Implementation-Robust Control1. 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?• …..

2. Functional robustness• Fault-tolerance towards computer-level faults leading to data errors• An increasing problem in future deep sub-micron technology

hardware


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)


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


Feedback Scheduling

• Dynamic on-line allocation of computing resources

• Feedback from actual resource utilization

• In principle, any computing resource


Feedback Scheduling Structures

• Feedback– Reactive

• Feedforward– Proactive

– Mode changes and admission control


Requirements on Scheduling Theory

• Relax the standard hard-real time assumptions

• Theory that better matches the needs of control systems


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


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


Jitterbug


TrueTime


Tool Usage

SchedulingParameters

(T,D,Prio, …)

Task TimingParameters

(latencies, jitter, …)

ControlPerformance

(variance, rise time, overshoot, ….)

Complex, ”nonlinear”relationship

Non-trivialrelationship

Simulation withTrueTime

Analysis withJitterbug


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)


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...


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


Possible SaabTech Issues

• Flexibility techniques for improved robustness• Combining – for mission-critical systems:

– Safe languages (Well-defined execution)– Certified run-time techniques (VM+HW)– Safe partitioning with shared resources.– Formal verification (FLEXCON+SAVE)– Improved testing techniques (pre-runtime)– Embedded on-line monitoring (run-time)

• Questions?

2004-11-23 @ SaabTech Jönköping FLEXCON Flexible Embedded Control Systems.

Documents

Transcript of 2004-11-23 @ SaabTech Jönköping FLEXCON Flexible Embedded Control Systems.