Dem@t-Factory (2010-2012)

New digitization chain

Supervising a system consists in the detection, isolation, identification and then repair of any fault that may occur in it. Nowadays, supervised systems are often distributed and communications are asynchronous.When communica- tions are disrupted, information takes an indefinite time to reach its destination, which slows down or stops the su- pervision process. The question is thus how to efficiently supervise a distributed system with unreliable communica- tions ? We claim that the answer to this question requires a distributed supervision system.In this document, we present ADS2, a multi-agent supervision architecture that is robust in a context of unreliable communications. ADS2 has four main characteristics :

  1. Each agent can perform both di- agnosis and repair,
  2. Each agent can find a balance be- tween a quick local diagnosis and repair, and a delayed, sys- temic one, based on the respective costs of misdiagnosis and communication,
  3. Agents may form a coalition to restore a consistent view of the system state in case some had to act locally with incomplete information at an earlier stage.
  4. Each agent is able to untertwine the diangostis and repair steps in order to reduce the size of the diagnosis search space and thus the time of malfunction.

The document develops our proposal along these four characteristics, and evaluates ADS2 using an industrial case-study.

Partners : Safig, A2iA, BancTec, Jouve, le laboratoire CEDRIC CNAM, le laboratoire A2iA ESIEE, LIP6, Temis

Associate Professor of Artificial Intelligence

My research interests include long-term autonomy, coordination, learning and decision making.