This poster presents ongoing work and mainly proposes a way to model goals on a Petri net model. We consider that the basic functioning of many machines can be captured in a Petri net model, while the environment where the machine is deployed is often too complex to be modelled in Petri nets. We handle the influence of the environment on the choice of operations by the so-called external agents , and show how these agents' goals could be studied before real deploym ent.
The basic idea of our approach is to use Petri nets [
PNSE’11 – Petri Nets and Software Engineering
In this work we are interested in finding a systematic approach for studying the behaviour of agents, based on their respective goals. Given a set of agents acting on a machine that is modelled in Petri net, how can we classify the behaviour of these agents? To the system designer, this approach is intended to give an understanding of the system prior to its implementation. Modelling the execution policy, that is, how the interaction between the agents and the Petri net model could be implemented is not addressed in this paper, but is shortly discussed in Section 3. 2
The five agents shown in Figure 2, are categorised according to four relations as illustrated in Table 1. The four relations are defined as follows: Distinctly Inclusive goals: We say that a goal ga distinctly includes gb if by achieving ga, gb is also achieved.
Mutually Inclusive goals: We say that two goals ga and gb are mutually inclusive if ga distinctly includes gb, and gb distinctly includes ga. Partially Inclusive goals: We say that ga partially include gb, when only some markings that achieve ga also achieve gb but not all of them.
Mutually Exclusive goals: We say that two goals ga and gb are mutually exclusive when they can not be achieved together. 3
In this paper we have addressed some aspects that are related to machines deployed in complex environments. We believe that the basic functioning of many
VIP g4 Air conditioning g5
Mutual Inclusion g2 g1 No No No
Distinct Inclusion g2 g1 No machines can be captured in a Petri net model, while the environments where the machines are deployed are often too complex to model in Petri nets. We consider that the influence of the environment on the machine is handled by agents, and raise the the following question: how agents acting on a machine can be categorised? We answer the question by introducing four properties, and use these properties to analyse some behavioural aspects, prior to system implementation as illustrated in Table 1.
We focused on the off-line analyses of agents acting on a machi ne, but we have not studied the on-line problematic, in other words the execution policy. When several agents are involved, which one should have the priority to execute? Another interesting question would be to find the set of necessary agents that guaranty some safety levels. To be more precise, could we provide a Petri net modelled machine with a set of safety agents, such that even in the presence of other faulty agents, the system guaranties the specified safety level? We believe that this work could be done within the Petri net formalism, and will be the subject of our future efforts.