This thesis presents sets of technologies for enacting multi-agent based decentralised workflow systems. Its purpose is to tackle some of the existing problems in the conventional workflow research from the system architectural and engineering point of view. Some of the problems addressed at the beginning of this thesis have affected the wide deployment of workflow management system in an open environment (internet). This thesis argues that most of these problems are caused by the huge conceptual gap and design rationale between high level application requirements and low level system design/implementation. Specifically, it is argued that the conventional system architecture of workflow management system (client-server) could be replaced by a multi-agent based platform which is more open, collaborative and can better reflect workflow's distributed features in the open environment.Resources:
Combining existing workflow design rationale and multi-agent computing technology, a multi-agent based decentralised workflow approach is proposed in this thesis. The architecture of the intended system removes both the centralised data storage and the centralised workflow engine from the system. To achieve this goal, approaches that bridge the gap between business process modelling and multi-agent interaction protocol production are proposed using three different techniques (namely functional properties based specifications verification, syntax based language mapping and interpretation based communication) according to the different types of business process models used. Based on such approaches, the mechanisms for decentralised process execution are explored. Moreover, our system is also able to be extended to support incompletely/particially specified processes in a distributed manner. The approach for handling such incomplete/particially specified processes at run-time are presented in this thesis.
The main contributions of this research is to provide approaches for enabling decentralised workflow systems in an open environment based on a multi-agent platform without changing the conventional workflow design rationale and with maximum use of existing process models and tools.