Article Preview
TopIntroduction
Nowadays, Multi Agent Systems (MAS for short) are omnipresent in computer science applications. This paradigm is used to resolve complex problems where reactivity, mobility, dynamicity and adaptation of the system to uncertain or unpredictable factors should be considered (Kouah, Saïdouni & Ilié, 2013). Then, the discipline of MAS engineering (i.e. specification, development, management and deployment … etc.) is becoming very important. It is concerned with models, methods and tools which are needed to develop MAS.
In the literature, several labors have been made-up to describe MAS, among others (Brazier, Dunin-Keplicz, Jennings & Treur, 1997; Caire et al., 2002; DeLoach, Wood & Sparkman, 2001; Demazeau, 1995; Ferber & Gutknecht, 1998; Jennings & Wooldridge, 1998; Fisher, 1994; Lind, 2001; Luck, Griffiths & d'Inverno, 1997; Marik, Müller & Pechoucek, 2003).However, they lack of an efficient methodology enabling designers to precisely built MAS with respect to its specification and dealing with almost MAS functionalities.
In fact, designing MAS needs both a specification model and an associated design methodology (Kouah et al., 2013). Concerning MAS specification models, several powerful formalisms have been proposed, such as Z-language (Regayeg, Kacem & Jmaiel, 2005), Petri nets (Celaya, Desrochers & Graves, 2007), colored Petri nets(ElFallah-Seghrouchni, Haddad & Mazouzi, 1999;Mazouzi, ElFallah-Seghrouchni & Haddad, 2002; Xu & Xie, 2011), recursive Petri nets (RPN for short) (Seghrouchni & Haddad, 1996), Maude (Mokhati, Boudiaf, Badri & Badri, 2007), Logic (Lomuscio & Sergot, 2003). As it has been clarified in (Kouah et al., 2013), these modeling approaches cannot deal with at least one of the following characteristics: abstraction and refinement, asynchronous aspects, synchronization between several processes. Accordingly, RPN has been extended to Synchronized Petri Nets model (SyPN for short). SyPN covers several functionalities of MAS such as: modeling abstraction and refinement, dynamicity, concurrency, preemption, recursion …etc.; enabling dynamic interactions between several Petri nets, preserving agents’ proprieties specially autonomy and enabling modeling of both internal and collective behaviors of MAS (Kouah et al.,2013).
Regarding design methodology, the formal refinement paradigm is an effective way to safeguard several design properties. In fact, abstracting something away is needed to deal with system constraints and designer limitations in gathering MAS requirements. Thus, refinement is needed to develop MAS in an incremental manner where details are added gradually.
At each abstraction level, a collection of abstraction (i.e. agents, interactions, processes …etc.) may be arbitrarily detailed and unambiguous but at the same time still contains abstractions that enable designer comprehension. So, MAS behaviors are not necessary well known at some abstraction level. Consequently, we introduce an unknown or undefined behavior which will be noted 
.