A Framework for Dynamically Configurable Embedded Controllers

1
A Framework for Dynamically Configurable Embedded
Controllers

• Johan Eker

2
Outline

• Introduction
• Pålsjö
• The PAL Language
• The Framework
• Patterns
• Case Studies
• Summary

3
Introduction

• What is an embedded controller ?
• How are they implemented ?
• Which are the problems ?
• What kind of tools do we need ?

4
Controller
Plant
5
Embedded Controllers

• The controller is embedded in the system.
• Parallel tasks.
• The control loop.
• User interaction.
• Logging and displaying data
• Real-time
• The computer must react to inputs from the plant
  at a certain speed.
• Deadlines, soft and hard.
• Large and complex.

6
Implementation Today

• Assembler, Forth
• C/C, Modula-2 real-time kernel
• ADA
• Dedicated systems
• PLC systems (SattLine).
• Matlab/X-Math with code generators

7
Which are the problems?

• C/C, Modula-2 or ADA
• Dynamic and portable.
• Long development time error prone.
• Code Generators
• Easy to use,but limited.
• Can be used together with simulation tools.
• Static configuration.
• General problems
• Hard to maintain and reuse.

8
What kind of tool did we need?
A rapid prototyping system that ...

• is easy to use and reuse,
• is portable, open and free,
• is on-line configurable,
• has supports algorithm libraries,
• and gives total control of execution.

9
and what did we get ? - Pålsjö!

• A framework for experiments in automatic control
  and embedded systems.
• A set of C-classes.
• Pålsjö Configuration Language - PCL.
• A language for control algorithm description.
• Pålsjö Algorithm Language - PAL Blomdell-97.

10
Outline

• Introduction
• Pålsjö
• The Pal Language
• The Framework
• Patterns
• Three New Patterns
• Case Studies
• Summary

11
Pålsjö ÈpùlÇSÎ
combines the advantages of the other approaches!

• Algorithms coded off-line.
• Block based language.
• Systems configured on-line.
• Creating, initializing, and connecting blocks.
• Data types tailored for automatic control.
• Matrices.
• Polynomials.

12

• Support for control algorithms.
• Periodic and sequential algorithms.
• A run-time system.
• User interface.
• Network interface.
• Support for several platforms.
• Open and free.
• Automatic documentation.
• Support for algorithm libraries.

13
Basic Ideas

• Observations
• Many controller implementations have a similar
  internal structure.
• The actual control algorithm is usually just a
  fraction of the total code.
• Solution
• Gather the common features in a library.
• Let the user only add the actual control
  algorithm.

14
The Pålsjö System
Function Libraries
User functions
PAL blocks
. . . . . . . .
Matrix functions
Polynomial function
15
On-line
Create blocks
Connect blocks
Real-time behavior
16
Outline

• Introduction
• Pålsjö
• The PAL Language
• The Framework
• Patterns
• Three New Patterns
• Case Studies
• Summary

17
PALPålsjö Algorithm Language

• Block based imperative language
• Signal and parameter interface
• Algorithm description
• Periodic and/or sequential.

Interface Description Algorithm Description
18
Interface Description

• Variables declared with a data type and an
  interface modifier.
• The interface modifier defines how the variable
  may interact with the environment
• Input, Output, State, Parameter
• name ltinterface modifiergt ltdata typegt

19
Data types

• Integers, floats.
• Strings
• Arrays
• Dimensions
• Integer variable used to specify dimensions for
  aggregate data types.

i input integer s parameter string a1
integer array1..3 of real a2 array0..2
of input real d 1 dimension a3
array1..d of integer
20
Data types

• Matrices
• Polynomials

d1 2 dimension, d2 2 dimension M
matrix1..d1,1..d2 of real deg1 3
dimension P1 output polynomial1..deg1 of
real P2 input polynomial1..deg12 of real
21
Periodic Algorithms

• Two special procedures
• Calculate() - Calculates output signals.
• Update() - Updates states.
• Example A PI-Controller

22
Module Module1 block PI y, yref, v input
real u output real K, Ti, Tr
parameter real P, I 0.0, e real h
sampling interval bi hK/Ti, br
h/Tr calculate begin e yref - y
P K e u P I end
calculate update begin I I bi
e br (v - u) end update end PI end
Module1.

Interface section

Algorithm section
23
Why two sweeps?

• Necessary in order to update states correctly
  when signals are limited.
• Keep the time delay from input to output at a
  minimum.

24
Sequential Algorithms

• Grafcet (IEC-1131)
• Textual representation for steps, actions and
  transitions

Initial step Init pulse activate Action1 end
Init step Step2 activate Action2 end Step2
action Action1 u . step Action1
transition from Init to Step2 when true
25
Outline

• Introduction
• Pålsjö
• The PAL Language
• The Framework
• Patterns
• Case Studies
• Summary

26
The Framework

• Contains classes for block administration, user
  interaction, network communication etc.
• A skeleton application.
• Main loop lies within the framework.

27
The Internal Structure

• Allows edits on a running system.
• Edits made on shadow configuration.

• Edits monitored and scanned for errors.
• States copied before swap.

28
Reflection

• Autonomous blocks
• Contain name, type, owner variable set info
• Functions for cloning and copying itself
• Autonomous variables
• Contain name, type, owner connections
• Functions for cloning and copying itself
• Plug-in objects
• String interface
• Data type will parse themselves
• C wrapper

29
Running the System

• PCL - Pålsjö Configuration Language
• Create, connect and execute blocks
• Reconfigure on the fly
• Text interface
• Network interface
• Setup
• Host - Target
• Single PC

30
The Interface
31
Outline

• Introduction
• Pålsjö
• The PAL Language
• The Framework
• Patterns
• Case Studies
• Summary

32
The Pattern Concept

• Package informal information.
• Document traditional solutions.
• Originates from architecture.
• Patterns
• A set of cooperating classes for a well defined
  problem.
• Problem definition, solution idea, participants,
  interaction,inheritance, sample code, known uses.

33
Three New Patterns

• The Calculate-Update pattern.
• Execution of block diagrams.
• The Parameter-Swap pattern.
• Assigning parameters to critical processes.
• The Register Idiom.
• Adding code to a framework without recompilation.

34
Parameter-Swap Pattern

• A real-time pattern that deals with the problem
  of updating controller parameters in a fast and
  consistent way.
• Scenario
• Several low-priority write processes
• One critical read process - the control loop.
• Design a structure which allows parameter
  assignment without blocking of the control loop.

35

• c is an indirect parameter.
• Calculation of indirect parameters may be
  expensive.
• Solution
• Use two parameter set.
• Let the writer process do all calculations.
• Use pointer-swap to change parameter set.

Block par in input real out output
real s state real a, b parameter real
c a b calculate out end
calculate update begin s end
update end par
36
The Run-time Behavior
37
The Classes

• The super classes encapsulated the relation
  between the block, the parameters and the
  environment.

38
Outline

• Introduction
• The PAL Language
• The Framework
• Patterns
• Case Studies
• Summary

39
Case Studies

• Inverted pendulum - standalone and on the robot.
• Code reuse and scaling.
• A hybrid tank controller
• Combining periodic and sequential algorithms.
• An adaptive servo controller
• Polynomials, external function and global
  dimensions
• A fault-tolerant scheme
• On-line configuration.

40
The Inverted Pendulum

• Swing up and stabilize.
• Equation of motion
• Measure the angle ?.
• Estimate the angular velocity.
  

41
The Controller

• Three sub-controllers
• Swing-up
• Catch
• Stabilize
• Good estimates for the angular velocity
  necessary!

42
A Nonlinear Observer

• The observer
• Analysis ? stability conditions.
• Design guidelines.

43
Experiments
44
Pendulum on the Robot
45
Outline

• Introduction
• Pålsjö
• The PAL Language
• The Framework
• Patterns
• Case Studies
• Summary

46
Summary

• Pålsjö
• A tool for real-time control experiments
• A new dedicated language - PAL
• NT, Solaris, m68k-VME, and PowerPC-VME.
• Future Plans
• New mechanisms for hybrid systems.
• Java GUI.
• Improved support for on-line algorithm update.
• Simulation support.

47
Simulation

• SIMART-A French simulation package for CACSD
• PAL blocks may execute in SIMART
• Experiments made with SIMULINK

48
Related Work
49
Contributions

• Design and implementation of the Pålsjö software
  environment.
• Three generic software patterns have been
  identified and documented.
• A novel nonlinear observer for the inverted
  pendulum.
• A new hybrid controller for a double tank system.

50
Papers and Reports

• Johan Eker and Karl Johan Åström (1996) -
• A Nonlinear Observer for the Inverted
  Pendulum.
• Johan Eker and Anders Blomdell (1996) -
• A Structured Interactive Approach to
  Embedded Control .
• Jörgen Malmborg and Johan Eker (1997) -
• Hybrid Control of a Double Tank System.
• Johan Eker and Anders Blomdell (1997) -
• Patterns in Embedded Control Systems.

51
Inheritance Structure
52
System Blocks

• The Periodic block
• Executes its child block periodically.
• Built-in scheduler.
• The Sporadic block
• Executes its child blocks sporadically.
• Useful for tasks such as interrupt handling.
• An Algorithm block must belong to a Periodic or a
  Sporadic block to be executed.

53
Example of a Configuration

• Down-sampled measurement signal.
• Two sampling rates.
• ? Use two Periodic blocks.

54
(No Transcript)
55
Simulations
56
A Hybrid Tank Controller

• Control the level of the lower tank.
• One controller for set-point changes.
• One controller for steady-state.

57
PID
Time Optimal
58

• Gain scheduled PID controller
• Time Optimal Controller
• Feedback using switching
  curves
• Switching PID ? Time Optimal using Lyapunov
  functions.

Level tank 2 (m)
Level tank 1(m)
59
Simulation
60
Experiments
61
Functions Procedures
procedure F(var1 input real var2 output
integer) begin . end
function F(var1 input real var2 output
integer) integer begin . result 1.0
end
62
A Typical Control System

• Several parallel tasks
• The control loop
• Read values from the plant,
• Calculate new control signals
• Write control signal to actuator.
• User interaction.
• Logging and displaying data