Delay Based SWIFI Approach to ABS

Dependability Analysis

Dawid Trawczynski, Janusz Sosnowski

Warsaw University of Technology

Institute of Computer Science

Nowowiejska 15/19, 00-665 Warsaw, Poland email:d.trawczynski@ii.pw.edu.pl

Outline

• Motivation• Problems in Dependability Analysis• Fault Injection Analysis Method• Fault Models• Delay Fault Model• Integrated Simulation Environment• Anti-Lock Braking Case Study• Summary

MotivationEmbedded systems:• make up a growing part of real-time

computing market• are reactive in nature and require

interaction with the environment• can support safety-critical

functionality

Problems in Dependability Analysis

• Fault set expansion and devevelopment of associated fault handling mechanisms

• Integration of modeling and fault simulation tools (various levels of abstraction)

• Selection of testing scenarios• Monitoring of system behavior that takes

into account interaction dynamics and running-time complexity

Fault Injection via Simulation

• Stress testing of various computing resources (ALU, FPU, cache memory, etc.)

• Hardware or Software• Software FI: highly contollable, allows

good observability and flexible - can emulate other faults types (e.g., hardware)

Fault Models

• Structural• Functional• Performance• Design • Malicious• Hybrid

Research Gap: Task Execution Delay Faults

Task Execution Delay Fault

∑=

=n

kks

1it

∑=

+=n

kkks

1

'i )(t φ

: fault-free task i execution time

: faulty task i execution time, where is the time perturbation random variable

Integrated Simulation Environment

Real-Time Simulator and Fault Injector integration:TrueTime + Matlab/Simulink + FITS + MSVC

Anti-Lock Braking System Case

Study

Distributed ABS Task Schedule

Experiments

• Single and four wheel ABS models• Tested a controller and a real-time network

via SWIFI• Injected TEF faults into controller and

network tasks • Measured performance (i.e., car braking

distance) versus the activation time of a TEF fault and the number of task deadline misses

Correlation of Performance, Task Deadline Misses and Time of Fault Susceptibility

X: lambda (interval of fault susceptibility)

Y: TDM (task deadline miss ratio)

Z: stopping distance (meters)

The Effect of TEF Faults on ABS Performance

Conclusion

• Enhancement of classical fault injection with abstract fault models

• Development of new test scenarios, temporal parameters, fault hardening mechanisms

• Assessment of fault effects via explicit or implicit approaches (e.g., performance, task deadline misses)

• Various phases of control algorithm can have different degrees of fault susceptibility

• Delay fault model can be applied to other time-critical applications (e.g., speech/image processing)

Thank You for Attention!