Disruption management in distributed enterprises: A multi-agent modelling and simulation of cooperative recovery behaviours A.C.A. Cauvin à , A.F.A. Ferrarini, E.T.E. Tranvouez LSIS UMR CNRS 6168—Paul Ce´zanne University, Domaine Universitaire de St Je ´ro ˆme, Avenue Escadrille Normandie-Niemen, 13 397 Marseille, Cedex 20, France article info Available online 21 June 2009 Keywords: Disruption management Actor relay Multi-agent modelling Simulation Networked enterprise abstract Disruption management in industrial areas consists in dealing with unanticipated events that get the plans deviate from their intended course. The solution results from the design and the maintenance of an operating mode ensuring a relevant deployment of individual recovery behaviours. The paper proposes an approach to minimize the impact of disrupting events on the whole system. It is based on an analysis of disrupting events and the characterization of the recovery process, and on a cooperative repair method for distributed industrial systems. This method is based on a cooperative distributed problem solving approach supported by a multi-agent system framework. & 2009 Elsevier B.V. All rights reserved. 1. Introduction: context and problematic In nowadays industrial context the intrinsic uncer- tainty of the environment generates events that disrupt pre-established plan controlling processes. The problem of disruption management is present in most of industrial areas and becomes a high interest topic (Yu and Qi, 2004). The main problem consists in dealing with these un- anticipated events that may disrupt the system and get the plan deviate from its intended course and even get it unfeasible. The problem becomes ‘‘how to reach the plan goals while minimizing the effects (negative impact) of the disruption(s)’’. In the case of distributed systems, this goal is implemented by the design and the maintenance of an operating mode ensuring minimal recovery time and consequences. This implies an efficient disturbance handling to ensure a quick and relevant deployment of individual recovery behaviours. A usual solution consists in pre-producing a recovery plan which integrates recovery solution(s) that can be applied on the day of the disruption. But this approach often lacks reactivity and decision-makers often stop after having generated a single feasible option for recovery; (computation time simply does not allow for several structurally different alternatives generation). However, (Clausen et al., 2001) , various works have relied upon distributed scheduling to generate more efficiently these plans. We distinguish four main classes of approaches among these works: Resource allocation, Decision support, Coordination and Cooperation (see Tranvouez, 2001 for a detailed review): The Resource Allocation approach consists in finding the most efficient resource. A detailed review is presented in Tharumarajah (2001). This process is repeated for all the tasks to be scheduled and is usually based on a call for proposal mechanism. The allocation can be controlled by a manager agent (named once for all or regularly changed) as in Rabelo and Camarinha-Matos (1998), decentralized in Ferrarini and Bertrand (1997) or mixed. Concerning the Decision Support approach, the solving process focuses on helping a human decision- maker as in Grabot et al. (1999). Such works intend to Contents lists available at ScienceDirect journal homepage: www.elsevier.com/locate/ijpe Int. J. Production Economics ARTICLE IN PRESS 0925-5273/$ - see front matter & 2009 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpe.2009.06.014 à Corresponding author. Tel.: +33 4 9105 60 61; fax: +33 4 9105 60 33. E-mail addresses: aline.cauvin@lsis.org (A.C.A. Cauvin), alain.ferrarini@lsis.org (A.F.A. Ferrarini), erwan.tranvouez@lsis.org (E.T.E. Tranvouez). Int. J. Production Economics 122 (2009) 429–439