Related Research

Owing to the increasing shortcomings of the traditional WfMS, many researchers in the related areas have invested a lot of time in enhancing the workflow management and proposed various modified workflow management system architectures. Researches about introducing the software agent technology into the workflow management systems have been invested for years. Yuhong Yan et al. [8]

got the related information about agent-oriented system architectures together and divide them into two ways. One of them is called agent-enhanced; its concept is similar to treat agents as tools for workflow automation. That is, they could only do something helpful for running workflows, such as acting as middleware for invoked applications. Because the objective of these agents is automation, they don’t need to interact and communicate with each other. There always exists a workflow engine monitoring and controlling all these agents’ behaviors. So these agents are more like a piece of ordinary software.

The other way is called agent-based; an agent based system means one in which the key abstraction used is that of an agent. At this time, software agents replace the behaviors and actions of business processes in the whole system. They not only did something helpful for running workflows, but also fulfilled the entire business process of workflow enactment. That is, the process logic is embedded in these agents, and they take full responsibilities of this WfMS. In other words, it means that they have all the means to analyze, automate, integrate and inspect workflows. Every agent is an independent individual and would have the ability to interact and communicate with

each other to accomplish the task. The software agent’s high-level capabilities, such as learning, negotiation and mobility etc., may be also put into the system architecture considerations. Although some abilities of them are not mature today, but it is obvious to see that the WfMS with these high-level abilities would be more flexible and powerful.

In this thesis, we introduce a software agent technology for WfMSs and describe the agent-based WfMS architecture to cope with the fast increments on open systems.

Firstly, an agent itself is an active object which runs in a loosely coupled distributed computing environment. Adding more agents and using the corporation mechanism of agents can solve the problems of scalability with large number of participant involvement. Secondly, when integrating with other applications and systems, the communication mechanisms and interfaces of agents provide more flexibility and extensibility than API calls. Thirdly, the interactions within a workflow may be much complicated and happen frequently. By making use of the agent communication languages such as FIPA ACL [18], SOAP [19] and KQML [20], the standards can be normalized with various interactions in a general form. Moreover, the autonomy, intelligence and pro-activeness of agents can provide personalized interfaces and timely help for workflow participants. Fourthly, the autonomy characteristic of agents makes an agent able to execute its own tasks or fulfill activities automatically.

Many research projects of agent orient workflow system have been presented and tried to fully use the agent capabilities in every dimension. One example is the TRP support environment (TSE) [21] proposed by Andersen Consulting, which has a central workflow authoritarian agent acting as a workflow engine and other authoritarian agents, actor agents etc. Its goal is to enable active collaborative work among participants workings on the TRP’s component based software engineering

environment. Another example is Advanced Decision Environment for Process Tasks (ADEPT) [7] project proposed by British Telecom Lab, which focuses on enhancing the supply chain management. The system consists of multiple software agents that concurrently negotiate an agreement on how resources should be assigned to support a business process. Later, Weishuai Yang et al. [9] proposed an agent-enhanced workflow (AEWF) model, which aims at enhancing the interoperability of workflow management engines. It is based on the BDI agent model, and tries to use mobile agents to increase the capability of workflow interoperability. Subsequently, Leangzhao Zeng et al. [10] proposed an approach that combines agents with workflows to effectively integrate cross-enterprise workflows. It is in order to support virtual enterprise and business-to-business e-commerce. Besides, although some projects among these researches claimed that they adopt the agent-based viewpoints, but according to the classification mentioned above, these projects are actually agent-enhanced systems.

These researches are all emphasized on inter-organization workflows, such as supply chain management and business-to-business. But they don’t really consider the significant problems happening to basic workflow definitions, such as the ones of scalability, extensibility, flexibility and adaptability mentioned in Chapter 1. Although, in order to solving conventional business-to-business (B2B) or business-to-customer (B2C) workflow behaviors, a sound service orient architecture has been proposed.

However, these workflows are almost very simple and stationary, they don’t even need to have a role model. A workflow becomes complicated just because of human involvement, such as cancel and countersignature. So here we primarily aim at presenting an agent-based system architecture for primitive workflows. And how to use agents to realize typical workflow behaviors such as start, stop, or enactment will

be described thoroughly in Chapter 4. On the other hand, these proposed systems still have fixed artifact representations only, which should be represented dynamically and adequately for varied process executors from current workflow’s viewpoints. Besides, the increasing facilitation not only occurs in inter-organization workflows, but also happens to workflows inside an organization. Thereupon, all of them should be solved effectively and flexibly in a modern workflow management system.