CONTEXT-AWARE DESIGN FOR PROCESS FLEXIBILITY AND ADAPTATION

ABSTRACT

Today’s organizations face continuous and unprecedented changes in their business environment. Traditional process design tools tend to be inflexible and can only support rigidly defined processes (e.g., order processing in the supply chain). This considerably restricts their real-world applications value, especially in the dynamic and changing environment. Recently, studies have contributed to the development of Adaptive Process Management Systems (APMS) that can facilitate fast implementation and deployment of business processes that allow for flexible adaptation. Most of them, however, only support ad hoc changes that require human intervention and only ensure structural correctness of these changes. This type of adaptation is time and effort consuming if such changes happen frequently. It is also difficult to trace and manage the sources of such changes so as to support automatic adaptation and improve reusability on the basis of past experiences. Furthermore, these approaches are not feasible in a knowledge-intensive environment because few of them examine semantic correctness of process changes. Hence, there is a need for new approaches that aim to design processes for flexibility and adaptability.

This dissertation introduces a new approach toward integrating context-awareness in process flexibility and adaptation to overcome these limitations. This dissertation centers around three main projects using principles of the design science methodology. First, we discuss the need of contextawareness in flexible process design and develop a formal approach to enable this design process. We propose to use an ontology-based method to model process contexts and Complex Event Processing (CEP) to detect critical situations. We also discuss the architectural support for context management.

Next, we propose various adaptation strategies and integrate them at both process model and instance levels in a context-aware manner. Specifically, we developed a process template and rule-based approach, considering business contexts such organizational policies, to configure process models at design time. This can handle a larger number of process models with small variance and facilitate their management when business objectives change. At the instance level, we propose a placeholder-based approach to customize the subsequent workflow on the fly based on the dynamic contexts and case data at runtime. Since the context that impacts process instance adaptation is highly domain dependent, we describe this work in clinical settings in this dissertation. We proposed a framework called ConFlexFlow, and showed how flexible and adaptable clinical pathways can be designed taking into account medical knowledge in the form of rules and detailed contextual information to achieve a high quality outcome. These pathways are selected during workflow execution based on rules that encapsulate medical knowledge and the dynamic context at runtime. Thus, each process instance is customized to an individual patient case based on the patient data, resources availability, etc.

Third, we propose a Mixed Integer Programming (MIP) model to check the compliance of process models and to validate the semantic correctness of process adaptations. We propose a formal specification language to model semantic constraints of activities, including presence and dependency relationships, ordering sequences, role assignment and obligations. Then the compliance issue is formulated as an MIP problem. We propose three novel ideas: the notion of a degree of compliance of a process, the concepts of full and partial validity of change operations, and the idea of compliance by compensation. Based on the above ideas, we use mixed-integer programming to check: (a) the semantic compliance of a process model and its evolution; and (b) if the ad hoc changes made to running process instances are semantically valid or not. If not, we calculate the minimal set of compensation operations needed based on degree of non-compliance, and transform a non-compliant process into a compliant one. We show that this novel approach is more elegant and superior to a pure logic-based approach.

Throughout this dissertation, most of the examples are from the healthcare domain for illustration. Clinical workflow has been considered as a killing area for the application of process management technologies, due to its dynamic, complex and knowledge-intensive nature. We prove that our approach can improve the care quality by increasing the flexibility and adaptability of clinical pathways while ensuring its semantic correctness.

 

Chapter 1 Introduction and Problem Statement

A business process (or workflow) comprises a series of value-added activities, performed by their relevant roles (human or computer applications) to achieve a common business goal. A business can be viewed as a collection of processes, and the robustness of these processes to a large extent is a crucial determinant of the success of the business. For example, order fulfillment, car insurance claim processing, and clinical pathways are critical processes respectively in the supply chain, insurance and healthcare organizations. A process management system (or workflow management system) is the software tool used to support design and execution of business processes represented in formal modeling languages. It supports coordination of activities among various people or computer applications and aims to improve business operational efficiency.

Today’s organizations often face continuous and unprecedented changes in their business environment. Conventional process management systems tend to be inflexible (Müller et al. 2004) and can only support rigidly defined processes (e.g., order processing in supply chain). This restricts their value in real-world applications considerably, especially in a dynamic environment such as mobile computing. Recently, a number of studies have contributed to the development of Adaptive Process Management Systems (APMS) that can facilitate fast implementation and deployment of business processes and allow for their flexible adaptation. However, most of them only consider ad hoc changes that are initiated manually for exception handling (Chiu et al. 1999; Hagen and Alonso 2000; Müller et al. 2004). In most work, structural correctness such as deadlock (or infinite loop) and data inconsistency is validated before such change is applied (Rinderle et al. 2004).

These approaches present two major disadvantages. First, the changes are made in an ad hoc way and they need human intervention; thus it is time and effort consuming. It is impossible to trace the sources for such changes so as to support automatic process adaptation and improve the reusability based on past experiences. Second, the semantic correctness of such change is overlooked since they assume that users are aware of domain knowledge related constraints. They are not applicable in the knowledge-intensive environment like the healthcare domain where semantic constraints play a key role, e.g., drug interaction. Hence, there is a need for a comprehensive approach that supports formal process context modeling, allows context-aware process adaptation, and preserves its semantic correctness.

1.1. Problem Statement

A business process is “a series or network of value-added activities, performed by their relevant roles or participants, to purposefully achieve the common business goal” (Ko 2009). Figure 1.1 presents a typical process lifecycle as it is promoted in both research and practice. Process design is driven by business goals taking consideration of organizational and legal issues. It translates high-level business objectives into concrete models (probably informal) that can be easily understood. Based on this, the next phase uses formal modeling languages to create executable process models. Three most popular workflow modeling languages include Petri Nets (Peterson 1981), BPMN (OMG 2006), and BPEL (Kloppmann et al. 2005), with different focuses. Expressiveness and complexity of process models (e.g., conditional branches, and loops) are balanced taking consideration of case data and possible scenarios. After that, a process instance can be initiated and it may encounter anticipated or unanticipated exceptions during its execution. The execution of process instances are usually monitored so that they can be analyzed or processed by data mining techniques. The last phase involves process evaluation according to the specific metrics and the result can be used to improve or reengineer the original process model.

A variety of factors can trigger the need for process flexibility and adaptation. First, the adjustment in high level business objectives such as regulations and organization policies can lead to the changes in business processes (i.e., phase I). Second, small differences in case data present potential changes to the predefined process model (i.e., phase II), which leads to a large number of process models with minor variance (Hallerbach et al. 2010; Kumar and Yao 2012). Third, during execution of each process instance (i.e., phase III), exceptions can happen that require deviation from its reference model (Hagen and Alonso 2000; Luo et al. 2000).

 

Figure 1.1. Process lifecycle

Defining the characteristics of variations that give rise to the need for flexible processes is essential to understanding how this need affects the requirements for flexibility (Kumar and Narasipuram 2006). It is important that flexible business processes should be designed in such a way as to meet the demands of variations. We define process context as situational circumstances that can impact process design (e.g. legislation, business policy, culture, etc.) and the execution environment in which a process is embedded (e.g. case data, performance requirements, time, location, etc). Contextualization of business processes has been emphasized and discussed at the conceptual level (Rosemann et al. 2008). However, a formal approach toward context-aware design for process flexibility is lacking.

Furthermore, two conflicting goals need to be balanced – need for control and need for flexibility (van der Aalst et al. 2009a). Process flexibility and adaptability is controlled by structural and semantic constraints. Structural correctness (e.g., no deadlock or data inconsistency) ensures error-free execution of processes. On the other hand, semantic constraints, derived from business policies and domain knowledge, enforce their semantic compliance. Business processes must comply with externally imposed regulations (e.g., such as business protocols, legislation, long-term contracts, and quality norms) and internal policies within an organization (Goedertier and Vanthienen 2006). Semantic constrains are derived from domain knowledge. For example in health care, semantic constraints can include medical knowledge such as drug-drug interaction and interdependencies of medical tasks (Ly et al. 2008).

1.2. Background and Motivations

1.2.1. A scenario in clinical settings

Healthcare is considered as the killer application area for process management systems (Dadam et al. 2000), since clinical processes are dynamic, complex, and knowledge-intensive. Thus, we use a clinical example to illustrate the aforementioned problems and challenges from the perspective of a real world application. Figure 1.2 presents a simplified clinical pathway, adapted from Blaser et al. (2007), for proximal femoral fracture in BPMN notation (OMG 2006). To some extent this process model is well and rigidly defined. For example, after the patient is admitted (T1) and then examined (T2), depending upon the result of examination, she will either take imaging diagnosis (T5) or have another diagnosis (T3) followed by therapy (T4). T5 refers to a subprocess that consists of X-ray test (T5_1) and CT scan (T5_2) executed in sequence, as shown the dashed box. Later in the process, other well-defined tasks follow, and eventually lead to the end of this process.

Although the above clinical process is strictly defined, there are a number of stimuli or triggers that can cause changes to this model and its instances. Such triggers include regulations, guidelines, business policies, culture, and particularity as they relate to case data (i.e., individual patients). Table 1.1 provides several example scenarios for process changes along with their sources. S1 and S2 are from regulations and hospital policies when a hospital wants to improve its clinical performance. S3, S4, and S5 are related to the case data (e.g., a specific patient) of a process instance since some region should be customized to meet the need of individual cases. For example, an old patient may need additional tolerance test before any imaging test. S6 and S7 reflect dynamics in the execution environment of process instances, since the attributes of resources (e.g., availability, service quality) can change over time. The dynamics of these case data and environmental data can be obtained from the workflow engine and require adaptation of process instances for specific cases.

 

Table 1.1. Example contexts for changes in the clinical process

S#	Description of the scenario	Source	Impact	Acquisition
S1	A patient must be instructed before imaging diagnosis (T5). (new regulation)	Regulation and policy	Process models	Business rule base
S2	A patient must sign a consent form before operative treatment (T8). (new hospital policy)
S3	A patient with bacterial infections should be administered with amoxicillin or clindamycin.	Differing  case data	Individual process instance	Process/ workflow engine
S4	For patients older than 70, an additional tolerance test prior to operative treatment is required due to possible risks.
S5	A pregnant patient has to take an MRI test or a sonogram test, instead of CT test.
S6	MRI test and CT test are interchangeable in case any device is unavailable.	Dynamic environment
S7	In case an administrative person is not available, patient admission can be deferred before her discharge.

 

On the other hand, process adaptation is restricted and controlled by structural and semantic constraints. Structuralconstraints refer to the control of process execution at the structural level. For instance, by verifying the absence of deadlocks and inconsistent data in a process model at design time, an APMS can determine whether a process is structurally correct or not. This is necessary to guarantee error-free execution of a workflow both before and after making changes. Semantic constraints stem from domain specific requirements and express dependencies, incompatibilities, existence conditions between activities, etc. (Ly et al. 2008). As an example, such a constraint may state that: possible drug interaction between amoxicillin and oral contraceptives prohibits a patient from taking both medications within 5 days. Similar constraints are also required to ensure that the process models are compliant with policies and regulations as well. Table 1.2 provides several examples of semantic constraints. Hence, an APMS must guarantee that process adaptation does not violate both structural and semantic constraints.

Table 1.2. Example semantic constraints for the clinical process

S#	Description of the scenario	Source	Impact
S1	A patient hypersensitive to penicillin should not be administered with drug amoxicillin or given a shot of penicillin.	Medical guideline	Process models and instances
S2	A patient must not be administered Aspirin and Marcumar within 5 days to avoid possible interactions.
S3	A patient with a cardiac pacemaker should be prohibited from having MRI test.

 

1.2.2. Process flexibility

Table 1.3 presents an analysis of the properties in process flexibility according to their triggers (i.e., process context) and nature of impacts. The properties in process flexibility and adaptation is based on the hierarchy developed by (Regev et al. 2006). As illustrated in the above section, three types of process contexts are discussed: Type I (business objectives), Type II (complicated scenarios), and Type III (exceptions). They can impact business processes at different levels and thus present different properties to realize the specified flexibility. We categorize the impact into three levels in decreasing granularity: process goal, process model, and process instance.

Table 1.3. Properties and types of process flexibility and adaptation

Process flexibility and adaptation			Nature of impact
Properties (Regev et al. 2006)		Process goal	Process model	Process instance
Extent of  change	Incremental		X	X
	Revolutionary	X
Duration of change	Temporary			X
	Permanent	X	X
Swiftness of change	Immediate		X	X
	Deferred		X
Anticipation of change	Ad hoc		X	X
	Planned	X
Process context		Type I	Type II	Type III

 

Flexibility at the instance level usually deals with the deviation from a standard process model for handling exceptions (i.e., Type III) encountered at runtime. Flexibility by deviation is temporary and only affects the current process instance, while other instances derived from the same model remain unchanged. It usually requires immediate change in an ad hoc way since the exception is usually unanticipated. For example, in an emergency situation of handling a patient with fracture, it would be appropriate to delay patient admission until treatment or surgery is finished. The overall process model and its constituent tasks remain the same.

When the same deviation takes place in many process instances frequently, the process designer should consider incorporating it into the process model (i.e., Type II). Another reason is that events may occur during process execution that was not foreseen during process design. Flexibility at the process model level is permanent and incremental. The change can be applied immediately or deferred depending upon the scenario. Change at the process model level is more complicated since it will affect the related running process instances and different migration strategies are required. For example, a process instance might have already completed the task where the change should be carried out. This instance may continue according to the old process model or restart from the beginning, depending on different migration strategies.

When there are too many changes that need to be made to the process model so as to affect the goal of this process (i.e., Type I), the modeler needs to redesign the process model and use it to replace the old one. Change at the process goal level requires redesign of a process model, and discarding the old process model along with all its process instances. This belongs to process reengineering where the underlying process model should be redesigned. It is out of the scope of this dissertation to explore this issue. Our focus is on the process context that comprises of exceptional events at the instance level, and complicated scenario at the model level.

1.2.3. Context handling in contemporary approaches

Contemporary approaches and techniques on improving process flexibility describe strategies for how to handle flexibility requirements rather than catching the triggers for process adaptation. Thus, a causeeffect relationship is also neglected in their works. A number studies that surveys flexible workflow design approaches (Regev et al. 2006; Schonenberg et al. 2007) but they also focus on the adaptation strategies. Context is mentioned in some studies but is never treated as the first-class citizen.

Recently years have witnessed some innovative researches in context-aware workflow management.

For example, Ardissono et al. (2007) developed a framework for adapting activities based on context; Adams et al. (2006) described a service-oriented framework for implementing dynamic workflows. Both consider context as a trigger for workflow adaptation. Although the importance of context is mentioned in (Kumar and Narasipuram 2006; Ploesser et al. 2010; Rosemann and Recker 2006; Rosemann et al. 2008) for process adaptation and flexibility, there is a lack of research in this perspective. Most researches only study one type of contexts. For instance, Modafferi et al. (2005) examined the capability of modeling business logic that is sensitive depending on the users’ context. They extended existing process modeling languages to allow modeling context sensitive regions. In contrast, Rosemann et al. (2008) emphasized the importance of external contexts such as weather and location. A variety of contexts can interact with each other and have different impacts on process flexibility. Without a formal and systematic approach to model, collect, and manage process context, it is difficult to automate context-ware workflow adaptation.

Context-aware process design is an orthogonal research dimension to the contemporary approaches.

However, it is highly relevant and critical for improving the research area of adaptive process design.

1.2.4. Motivations

Traditionally and from the technical perspective, process management and context management are two independent research areas (Sell and Springer 2009). The research in the area of context management focuses on the design of context models, context services, sensor-based context acquisition, etc. while the study in adaptive process management concentrates on the adaptation strategies and merely regards the reactive part of process adaptation and not its actual trigger. Although researches have been increasingly interested in the impact of context in process flexibility and adaptation, few of them use a formal methodology to support process context modeling and management.

In summary, an adaptive process management system without context-aware capability is unable (1) to support automatic process adaptation when critical contexts are detected, (2) to track the cause-effect relationship between the trigger and the adaptation, and thus reduce the reusability of process fragments that are constructed at runtime. Despite the wide use of the context-aware approach in various domains, its use in process adaptation is limited. Researchers have only introduced this approach recently recognizing that in a dynamic and changing environment, business processes should be adaptable to the contextual information (e.g., unavailability of resources, change of policy). The major difference between our approach and contemporary works is the emphasis on “context”.

Another issue in managing context-aware workflow is the semantic correctness of processes change or adaptation. Although this is an important issue especially in a knowledge-intensive environment, it is not fully addressed in contemporary approaches. It should be considered because automatic adaptation without considering process compliance is not likely to be adopted. Thus, our study also aims to examine the compliance issue in context-aware workflow management.

1.3. Research Framework and Contributions

1.3.1. Research framework

Motivated by the above observations, this dissertation proposes a framework that allows contextaware process adaptation and meanwhile ensures its compliance, as shown in Figure 1.3. This framework also shows a number of research issues and groups them into the following three categories. Next, we discuss the research questions in each category and our technology-based approach to address them.

 

Figure 1.3. Research framework

Category I: Supporting context-aware capability in business processes

• Problem 1: How to apply context-aware approach to improve process flexibility and adaptation?
• Problem 2: How to model and manage (i.e., acquisition, reasoning, and dissemination) context?

Category II: Context-triggered process reconfiguration and adaptation

• Problem 3: How to support context-aware flexibility in process models (i.e., reconfiguration of process model based on context change)?
• Problem 4: How to support context-aware adaptation in process instances (i.e., adaptation in process instances based on context change)?

Category III: Ensuring the correctness of process reconfiguration and adaptation

• Problem 5: How to ensure structural correctness?
• Problem 6: How to ensure semantic correctness?

 

A number of studies have contributed to the research in adaptive process management, and most of them focus on the adaptation strategies for handling exceptions or ad hoc changes (problem 4), e.g.,

(Hagen and Alonso 2000; Luo et al. 2000), and their structural correctness (problem 5), e.g., (Müller et al.

2004; Rinderle et al. 2004). However, they only deal with adaptation strategies, i.e., how the adaptations are performed, rather than by what and when the adaptations are triggered (Sell and Springer 2009) (problem 2). Thus, no formal method is provided to support context-aware design for process flexibility and adaptation (problem 1).

In an dynamic and pervasive computing environment, where sensors and embedded systems are distributed, process adaptation can occur frequently and unmanageable (De Leoni et al. 2007). By monitoring workflow activities, behavior of workflow participants, and environmental factors, a business process can have the capability of sensing potential changes and making adaptations accordingly in a predictive manner. For example, the physiological data of a heart-attack patient is monitored and analyzed continuously during her treatment. Once any critical situation is detected, her subsequent treatment plan needs to be changed according to her current situation.

Thus, we need a formal methodology to contextualize business processes and formally model these contexts. Meanwhile, context-triggered process adaptation requires less interruption from human being, which is more likely to cause violation of semantic constraints. Thereby, providing adaptive processes in an automatic and predictive way should take into account compliance checking of the possible adaptation patterns against semantic constraints (problem 6), especially in knowledge-intensive applications.

1.3.2. Contributions

To address the above issues and problems, this dissertation focuses on how to achieve contextawareness in process flexibility and adaptation while ensuring their semantic correctness, following the design science methodology. Figure 1.4 summarizes the abovementioned research issues and proposes our solutions organized in chapters. It includes the following five components:

• Context-aware design approach: this part uses a formal methodology by extending traditional context-aware design approach for process flexibility and adaptation. The standard approach includes three parts: context specification, management, and usage (or action). Our focus is context usage that associates context change with context-triggered adaptation behavior. Further, we add validation and reuse to the standard process to guarantee the action (i.e., adaptation patterns) is semantically correct and can be reused for future analysis. Such cause-effect relationship between context and adaptation patterns is helpful to improve the current process model.
• Context specification and management: process context can come from heterogeneous sources that use different data formats and semantics. They can be obtained by referencing related documents, communicating with domain experts, or receiving low level events from sensors. To facilitate knowledge sharing and communication, we use ontology-based approach to model process context with reasoning capability. Further, in a pervasive computing environment where data streams is in high-volume and high-speed, distributed data should be correlated in a timely fashion to deliver actionable information. We use Complex Event Processing (CEP) to detect critical situations.
• Context-triggered process reconfiguration: a process model should adapt to context change in an intelligent way. We use a template and rule-based approach to configure process models on the fly based on the available context. This approach can reduce the large number of process models (a.k.a. process variants) and efforts involved in maintaining these models when a single policy changes, because it allows separation of basic process flow from business policy elements in the design of a process and also integrates resource and data needs of a process tightly. We also developed a novel scheme for storing process variants as strings based on a post-order traversal of a process tree. We showed that such a representation lends itself well to manipulation and also for searching a repository of process variants.
• Context-aware process adaptation: a process instance should also adapt to context change in an intelligent way. A process execution engine should be notified of context change and support a variety of adaptation strategies according to different scenarios. We use the placeholder activity (i.e., ad hoc subprocess) to handle planned changes with dynamics at runtime, and illustrate the application of this approach in clinical settings. We showed how adaptable clinical pathways can be designed taking into account medical knowledge in the form of rules and detailed contextual information to achieve a high quality outcome. These pathways are selected during workflow execution based on rules that encapsulate medical knowledge and various clinical contexts.
• Semantic correctness: any adaptation to be applied to a business process should be validated against its structural and semantic constraints. Since structural correctness (problem 5) has been extensively studied in (Rinderle et al. 2004), we only focus on the verification of semantic constraints, which stem from domain specific requirements and express dependencies, incompatibilities, and existence conditions between activities. We propose a formal constraint specification language to model semantic constraints. Further, we apply Mixed Integer Programming (MIP) to check the compliance of processes and validity of process changes during their lifecycle.

 

*Note: Problem 5 has been well handled by other studies, so it is excluded from our research issues

Figure 1.4. Research issues and solutions organized by chapters

 

The major objective of our study is to improve the flexibility and adaptability of business processes while ensuring their semantic compliance so as to make it applicable in dynamic and knowledge-intensive areas. The contributions of this study are many-fold:

• We propose a formal approach that integrates context-awareness in adaptive process design. Our approach complements related works in the BPM community from the perspective of process contextualization and context modeling. Instead of focusing on the adaptation strategies to support ad hoc changes, our study shifts the emphasis to the triggers that have a potential impact on process design and execution. Thus, it does not require human intervention and facilitates reuse of adaptation patterns.
• We developed an ontology-based context model for health care to capture important concepts (or contexts) and their relationships in clinical settings. With semantic web technologies, this model improves interoperability among distributed software systems and embedded devices. Thus, we address the issue of heterogeneity in healthcare information systems and captures contexts for clinical workflows.
• We use CEP technology to detect contexts that are composed by basic events in an event-driven architecture. CEP demonstrates an efficient capability in modeling and correlating hospital events in large volume and high speed. We show that our approach is efficient in processing real-time events and detect critical situations in a timely fashion.
• We propose two types of context-aware adaptation behavior both at the model and instance levels. First, we use a template and rule-based approach to configure process models at design time to improve process flexibility. It can balance the complexity and the number of process variants in a large repository. Second, we propose to materialize a placeholder activity at runtime for adaption in the context of clinical settings. We show that clinical pathways are selected during workflow execution based on rules that encapsulate medical knowledge and various clinical contexts that capture different patient cases. Both approaches are triggered by context change and do not require human intervention.
• Finally, we propose a formal specification language to model semantic constraints and use a MIPbased approach to model compliance checking of process models and semantic correctness of change operations to be applied in process instances. Our approach is novel and presents advantages over pure logic-based approaches.

1.3.3. Organization of the dissertation

This dissertation is organized as follows. Chapter 2 surveys related works in process flexibility. It provides a comprehensive review of approaches in achieving process flexibility and the context handling in these studies. It exposes a major research gap in this area, which motivates us to do this study. We also explore the use of workflow technologies in handling healthcare processes. Then Chapter 3 extends a context-aware design approach to model and manage process context in a formal way. Following that approach, we discuss context specification and detection in Chapter 4. Further, context-triggered process model configuration is handled in Chapter 5, which introduce our template and rule-based approach. Chapter 6 introduces context-aware process instance adaptation and illustrates this approach in clinical settings. Then, Chapter 7 describes an MIP-based approach for checking the compliance of process models and the semantic correctness of change operations during process lifecycle. Finally, Chapter 8 concludes this dissertation and presents our future work.

CONTEXT-AWARE DESIGN FOR PROCESS FLEXIBILITY AND ADAPTATION

Sharing is caring!