عامل کار : یک سیستم جریان کار حمایت از قاعده مبتنی بر انطباق جریان کار

Current workflow management systems still lack support for dynamic and automatic workflow adaptations. However, this functionality is a major requirement for next–generation workflow systems to provide sufficient flexibility to cope with unexpected failure events. We present the concepts and implementation of AgentWork, a workflow management system supporting automated workflow adaptations in a comprehensive way. A rule-based approach is followed to specify exceptions and necessary workflow adaptations. AgentWork uses temporal estimates to determine which remaining parts of running workflows are affected by an exception and is able to predictively perform suitable adaptations. This helps to ensure that necessary adaptations are performed in time with minimal user interaction which is especially valuable in complex applications such as for medical treatments.

Workflow management is widely adopted as a core technology to support long-term application processes in heterogeneous and distributed environments [2], [17] and [21]. Main characteristics include the clear separation of application program code from the overall process logic and the integration of automated and manual activities. Workflow technology is increasingly used to manage complex processes in Internet-based e-commerce, virtual enterprises, or medical institutions [33] and [49]. For example, due to precisely specified treatment procedures in many medical disciplines, workflow management systems can be used to implement diagnostic and therapeutic processes [13], [40] and [49]. Major goals include the improved and timely treatment of patients and a significant workload reduction for the hospital personnel. However, conventional workflow management systems do not provide sufficient flexibility to cope with the broad range of failures that may occur during workflow execution. In particular, not only system failures such as hardware or software crashes need to be dealt with but also logical failures or exceptions. These logical failures refer to application-specific exceptional events for which the control and data flow of a workflow is not adequate anymore and thus has to be adapted [54]. The automatic treatment of such logical failures is the main subject of this paper. In the cancer chemotherapy workflow shown in Fig. 1, assume it is detected just before the administration of drug C that the leukocyte count (i.e., the number of white blood cells) has become critically low, so that there is the risk of a serious infection for the patient. As drug C is known to reduce the leukocyte count additionally as a negative side effect, the activity “Administer drug C” dynamically has to be removed from the workflow while the execution of the other activities can be continued without change. To protect the patient from an infection, it may also be necessary to dynamically add an activity supporting the administration of an antibiotic drug after the cancer chemotherapy. Note that explicit conditional routing paths in the workflow definition are not sufficient to deal with such exceptions. For example, checking the condition “leukocyte count < 1000” before the “Administer drug C” activity would not help if this condition is violated at different points in time possibly requiring different actions (e.g., dropping drug A instead of drug C). Inserting conditional branches at any potentially relevant position would significantly reduce workflow readability and maintainability. Thus a more flexible exception handling is required to decide on how to best react to logical failures.Previous work on dynamic workflow adaptation mostly focused on a manual approach where the administrator or an authorized user has to decide which events constitute logical failures and which adaptations have to be performed [45]. However, the manual approach can be time-consuming and error-prone thereby threatening the goals to be achieved with workflow management. For example, during a therapy such as the one shown in Fig. 1, a physician is usually faced with up to 20 patients and 10–30 findings per patient every day. With a manual failure handling, the physician always would have to keep in mind which findings may induce which adaptations, or at least would have to look it up in text books in a time-consuming manner. Hence, events constituting logical failures may be overseen or detected too late. Recent approaches supporting automated workflow adaptation typically limit adaptations to the currently executed workflow activities [7] and [9]. Such an approach is only of limited usefulness as all workflow parts not yet reached by the control flow are not adapted automatically. This may also lead to situations where necessary adaptations are performed too late so that significant problems can occur. For example, adding a new drug administration in a cancer therapy typically requires ordering the necessary drugs one or two days before the scheduled administration to prepare a patient-specific infusion. Thus, in order to allow a timely drug administration the corresponding workflow adaptation should be performed as soon as possible. Similarly, the dropping of a cancer drug (such as drug C in Fig. 1) should not be performed in a “last minute” manner but in advance to avoid that a very expensive drug infusion has to be poured away. Of course, early scheduling of new activities and avoiding the unnecessary execution of originally planned activities are of great importance in many workflow application domains, e.g., for product delivery in supply chain management and writing reviews in evaluation processes. To overcome the limitations of existing systems and comprehensively support automated workflow adaptations, we designed and developed the workflow management prototype AgentWork. It is the first system we know of that can predictively adapt the yet unexecuted parts of running workflows in a largely automated manner. The implementation of such a capability poses many challenges, in particular support for a temporal model in the specification and treatment of logical failures. This paper gives an overview of AgentWork and its underlying concepts. The contributions of our work are as follows: • We support two strategies for automatic workflow adaptation called reactive and predictive adaptation. Predictive adaptation adapts workflow parts affected by a logical failure in advance (predictively) based on temporal estimates of the affected workflow activities. The adaptation typically takes place as soon as the failure is detected thereby often providing enough time to meet organizational constraints for adapted workflow parts. Reactive adaptation is performed when predictive adaptation is not possible. In this case, adaptation is performed when the affected workflow part is to be executed. In particular, before an activity is executed it is checked whether it is subject to a workflow adaptation such as dropping, postponement or replacement. We provide mechanisms to decide whether reactive or predictive adaptation is more suitable for a particular failure situation. • We provide an ECA (Event/Condition/Action) rule model to automatically detect logical failures and to determine the necessary workflow adaptations. To support predictive workflow adaptations, we use a temporal object-oriented logic that allows us to specify the valid time interval for which an adaptation has to be performed. Furthermore, our approach supports the integrity of ECA rule sets. • We provide workflow estimation algorithms to determine which workflow part is affected by a logical failure and needs to be adapted. • We support a comprehensive set of operators for automatic workflow adaptation, including control flow operators which for example allow us to add or delete workflow activities. Furthermore, data flow operators are provided that adapt the data flow after a control flow adaptation, if necessary. • Finally, we provide mechanisms to monitor adapted workflows by checking whether the used time estimates are met when the adapted workflow is continued. As a first application area, AgentWork supports workflows for cancer treatment in an interdisciplinary medical project at the University of Leipzig [39] and [40]. Though important conceptual decisions are motivated by this medical workflow application, AgentWork has been designed to be usable in other workflow application domains as well (such as insurance business or banking). In particular, the basic AgentWork model only assumes generic events and workflow activities. By sub-classing, these generic events and activities can be refined in a domain-specific manner (e.g., for a business domain) without affecting the workflow adaptation model. The paper is organized as follows. In the next section, we give an overview of the AgentWork system. Section 3 describes our ECA rule model. Section 4 presents the approaches for selecting the adaptation strategy, workflow duration estimation, control and data flow adaptation, and workflow monitoring. Finally, we discuss related work (Section 5), and summarize and sketch future work (Section 6).