Eric Dallal

About

I am a 5th year Ph.D. student working with Professor St├ęphane Lafortune at the University of Michigan in Ann Arbor. My work is in the field of Discrete Event Systems (DES). I have two current research projects: dynamic fault diagnosis and DES abstractions for vehicle control. I am also working on the DESUMA 2.0 program, a graphical software system built on JGraph for performing the DES operations of the UMDES toolbox.

Research

Dynamic Fault Diagnosis

This work is concerned with the diagnosis of DES modeled by finite state automata (FSA) under dynamic partial observation. The goal in this work is to diagnose the occurrence of an unobservable fault event given a subset of events for which we have sensors (called monitorable), while minimizing sensor activations. We define a structure called the Most Permissive Observer (MPO), first introduced in Fault Diagnosis with Static and Dynamic Observers, which contains all sensor activation policies that satisfy the property of K-diagnosability. By mapping the problem to that of state disambiguation and making use of monotonicity properties, we can construct a more compact version of the MPO, which reduces computational complexity.

DES Abstractions for Vehicle Control

This work uses abstraction and DES methods to provably ensure safety of vehicles crossing an intersection. We assume sources of uncertainty in the form of uncontrollable vehicles and unmodeled dynamics and seek to construct a supervisor satisfying the following three properties:

  • Safety: The vehicles must not collide.
  • Non-blockingness: The vehicles must eventually cross the intersection, never reaching a state where the supervisor's set of allowable actions is empty.
  • Maximal Permissiveness: The supervisor must allow any action that does not cause of violation of safety and non-blockingness.
We begin by constructing an abstraction of the continuous time system by discretizing the system in space and time, and proceed to define uncontrollable events to model the sources of uncertainty. We then make use of DES methods to obtain a supervisor that provably satisfies the three properties listed above.

Projects

DESUMA

DESUMA is a graphical software system used to perform various operations on DES, such as composition, control, and diagnosis. It is made up of a Java front-end for the graphical user interface (GUI) and a collection of C utilities from the UMDES toolbox. The current version available is built on the Grappa graph visualization package for Java. Because of this, DESUMA currently works only with older versions of Java. Development has been ongoing over the past few years to replace Grappa with JGraph so as to allow DESUMA to run with the newest versions of Java. This work, being conducted at the University of Michigan and at Mount Allison University, is now nearing completion.