=Paper=
{{Paper
|id=None
|storemode=property
|title=Smart Process Management: Automated Generation of Adaptive Cases based on Intelligent Planning Technologies
|pdfUrl=https://ceur-ws.org/Vol-615/paper10.pdf
|volume=Vol-615
|dblpUrl=https://dblp.org/rec/conf/bpm/Gonzalez-Ferrer10
}}
==Smart Process Management: Automated Generation of Adaptive Cases based on Intelligent Planning Technologies==
https://ceur-ws.org/Vol-615/paper10.pdf
Smart Process Management: Automated Generation of
Adaptive Cases based on Intelligent Planning
Technologies?
Arturo González-Ferrer, Juan Fdez-Olivares, Inmaculada Sánchez-Garzón, and Luis
Castillo
Department of Computer Science and Artificial Intelligence, Universidad de Granada,
{arturogf,faro,isanchez,l.castillo}@decsai.ugr.es
Abstract. This paper presents a proposal for the seamless integration of Intelli-
gent Planning techniques into the life-cycle of BPM. The integration is intended
to leverage current BPM techniques by allowing them to manage smart processes
as adaptive business cases that can be automatically generated from original pro-
cess models and executed in standard BPM runtime engines. The integration of
such intelligent techniques is based on a two-fold transformation process: from
business models into planning domains, and from plan representations into exe-
cutable processes.
1 Motivation
Adaptive Case Management [15] is being used by the Workflow Management Coalition
as the brand new name of an emergent paradigm in current BPM standard aimed at sup-
porting Human-Centric processes[5] for knowledge workers (highly qualified personnel
of organizations, like experts or decision makers). The processes required by knowl-
edge workers are collections of tasks, which usually are collaboratively performed and
which necessarily require human interaction in order to control and manage their exe-
cution. Such processes commonly support decisions and help to the accomplishment of
workflow tasks of knowledge workers in several application domains. For the sake of
simplicity, we will designate these processes as Smart Processes.
Smart Processes may be viewed as business cases that demand some kind of in-
telligent management [15] since, on the one hand they are very difficult to foresee, as
they respond to unstructured sets of procedures which reside either in experts’mind or
in documents, what makes difficult to devise a priori which tasks to execute. On the
other hand, they need to be adaptively generated, since they are unpredictable and they
strongly depend on the context of the organization and do not respond to a fixed pat-
tern. Finally, they have to be flexibly and interactively executed by humans since they
are subject to change.
There is a general consensus on that BPM technologies should be improved in or-
der to support this kind of processes since, at the time being, BPM is mainly focused on
?
This work has been partially supported by the Andalusian Regional Ministry of Innovation
under project P08-TIC-3572.
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
the management of static, repetitive, even perfectly predictable tasks/processes, mostly
devoted to low qualification operators[15]. This is a widely known weakness and, be-
cause of this, it is also recognized that new techniques must be developed at both steps,
process modeling/generation and process execution, in order to fully cover the needs of
knowledge workers on Smart Processes.
In this sense, we present in this work a proposal that leverages the current BPM
life-cycle in order to support smart processes through the development of Knowledge
Engineering and intelligent planning techniques, focused on a two-fold transformation
process. On the one hand, a transformation from business models into planning do-
mains, in order to make the output of a business modeling tool interpretable by an
intelligent planner. On the other hand, a transformation from a plan representation into
an executable process, in order to make the output of the planner understandable by a
BPM runtime engine.
The reason for the first transformation process is the fact that Artificial Intelligent
Planning and Scheduling [10] AIP&S is a technology that clearly covers the above
exposed demands for smart processes. Concretely, the hierarchical planning paradigm
(mostly based on HTN, Hierarchical Task Networks [14,4]) has been proven in many
applications([6,3,2,7]) to be successful on supporting the knowledge workers’ effort.
On the one hand, by modeling expert knowledge with planning domain models (which
allow the description of actuation protocols or operating procedures represented as a
hierarchy of tasks networks [13,4]). On the other hand, helping them to adaptively
produce plans to support their decisions, as the result of a knowledge-driven planning
process, guided by the knowledge represented in the planning domain. The reason for
the second transformation process is that BPM has demonstrated to be much more ap-
propriate for supporting the execution of the result of knowledge workers’ effort, by
providing technological infrastructures in order to interactively execute and monitor
processes. Therefore, translating a generated plan into a BPM executable format will
allow to execute plans on already existing standard platforms. Both transformations are
fully automated, what allows to seamlessly integrate these techniques into the current
BPM life-cycle, leading to an integrated framework for Smart Process Management that
supports the automated generation of adaptive cases, from an original business process
model based on AI P&S techniques.
Next sections are devoted to briefly explain the most relevant aspects of the frame-
work as well as its main advantages.
2 Integrating Intelligent Planning into the BPM life-cycle
AI Planning and Scheduling [10] and more concretely HTN planning [4,13] becomes
the central technique for this work since it supports the modeling of planning domains
in very similar terms to the ones used in standard BPM models. An HTN domain is
a compositional hierarchy of tasks networks representing activities at different levels
of abstraction (either compound or primitive tasks). A domain describes how every
compound task may be decomposed into (compound/primitive) sub-tasks and the order
that they must follow, by using different methods. An HTN planner interprets the set of
task decomposition schemes and reasons about them in order to compose a suitable plan
(a set of activities subject to order and temporal relations) such that its execution reaches
a given goal, starting from an initial state (that represents an initial context as initial
values for the properties of objects or resources involved in the activities). The HTN
reasoning process is a knowledge and goal-driven process, guided by the procedural
knowledge encoded in the domain. HTN techniques have been recently enhanced with
valuable temporal and resource reasoning extensions [4], what allows to cope with a
very rich temporal and resource representation, as well as to obtain plans that could
be flexibly executed since they contain temporal constraints that can be adapted during
plan execution.
From the AI P&S point of view, the need to obtain a context dependent, concrete
workflow from a given business process model can be seen as the problem of obtaining
a situated plan from (1) a planning domain that represents the original process model
and (2) from an initial state that represents the context for which the business case has
to be adapted. This aspect is the cornerstone of the proposal here presented, called Jab-
bah, a Knowledge Engineering for Planning tool that supports a three-step process that
starts on an initial, already defined business process model, represented in XPDL us-
ing standard BPM modeling techniques. At a first step, the XPDL-based process model
is transformed into an HTN planning domain and problem. Second, an HTN planner
taking as input the domain and the problem (representing the context under which the
case has to be enacted), generates a situated plan that represents the case to be executed.
Hence, by using Jabbah in order to generate HTN domain and problem files, from an
original process model, it is possible to carry out a knowledge-driven HTN planning
process that results in the generation of situated plans, that is, plans customized for a
given situation. These plans can be used either for supporting decision making about ac-
tivity planning or process validation based on use-case analysis, leveraging the current
BPM life-cycle at its process modeling/generation step. Third, the plan is finally trans-
formed into a process in an executable format (XPDL again), and this process is then
used as input to a standard BPM runtime engine in charge of supporting its interactive,
human-centered execution. Details about these steps are explained in the following.
Transformation from process models to planning domains. Given an XPDL pro-
cess as input (which can be clearly seen as a graph), Jabbah proceeds by identifying
common workflow patterns (that is, sequential, parallel, subprocess and conditional
structures) as process blocks in the process model, and then generate a tree-like struc-
ture, much similar to HTN domains. The HTN target domain language (called HTN-
PDDL) used in this work is a temporally extended, hierarchical extension of PDDL
[9], the standard language of planning domains (see [4,7] for details about this repre-
sentation). Concerning the representation of preconditions and effects found in plan-
ning representations, we only deal at the moment with the conditions that have been
defined in the original BPMN model, as well as the task order established on it. How-
ever, the BPMN notation allows the inclusion of customized annotations, that could be
used to augment the knowledge about preconditions and effects present on the process
model (i.e. by using extendedAttributes for activity nodes). This said, the Knowledge
Engineering process for transforming process models into planning domains consists of
three different stages: i) Firstly, the XPDL document is parsed, transforming it into an
intermediate data structure and graph model that can be easily managed throughout the
next stages. ii) Then, the different blocks of workflow patterns (serial, parallel, subpro-
cess and conditional blocks) are detected, distinguishing their kind from the knowledge
acquired in the previous parsing stage, and build up an equivalent tree-like model. To do
this, a collateral challenge is the transformation of the graph into a tree-like structure,
which has been done using an algorithm described in [1]. This is carried on by arrang-
ing those workflow patterns hierarchically, but also keeping the semantic information
(about control flow and decisions) present in the process diagram (see [11] for more
details). c) Finally, we need to do a planning language generation phase, where we an-
alyze the tree model that has been populated previously, trying to generalize common
patterns found in the graph (i.e. serial or parallel split-joins patterns are always coded
in the same way), and writing the HTN-PDDL code that corresponds to the tree-graph
fragment analyzed.
(a) (b)
Fig. 1: A well-structured process model is designed with a BPM suite (a), and trans-
formed later into a corresponding tree model by Jabbah (b), more appropriate to extract
the HTN planning domain
By following this process, it is possible to generate domain and problem files which
are given as input to an state-of-art HTN planner in order to obtain situated plans. We
TM
have used the IACTIVE planner for this work, a temporally extended HTN planner
which uses HTN-PDDL as its planning language. Moreover, it has already been used
in several applications [3,7,6]. These plans are generated by the planner for a given
context represented in the problem file, and they can be interpreted as adaptive business
cases since they are direct and automatically obtained from the initial process model.
Given that the context parameters that guide the deliberative reasoning of the planning
stage are included in the problem file, dynamic changes on the environment should
be monitored, modifying the problem file accordingly, triggering a replanning stage
to generate a new situated plan (the domain file would not be modified, in order to
respect the original process model). Some approaches already exists for the monitored
execution of plans [12], in order to handle any kind of exogenous events. The design of
such execution monitor for the IACTIVE planner is being carried out at the moment of
writing this paper.
Next, we briefly describe how these plans are transformed back into XPDL process
instances in order to be understandable, and so executable, by a BPM engine.
Transformation from plans to executable process models. Given an XPDL pro-
cess instance as input, BPM engines are commonly endowed with the necessary ma-
chinery in order to interactively execute every task in the process (allowing to start,
finish, suspend or abort it) by following an execution model based on state-based au-
tomata. The plans generated by the planner, using the planning domains and problems
generated by Jabbah, are represented in XML as a collection of Task nodes where ev-
ery node contains information about: actions (activities) and their parameters; temporal
information as earliest start and earliest end dates for the execution of every activity;
order dependencies between actions which allow to establish sequential and parallel
runtime control structures; and metadata which allow to represent additional knowl-
edge like the user-friendly description of a task, its type (manual, auto) or its performer
(that is, the participant of the activity). It is worth to note that metadata are generated
at domain generation phase and are automatically extracted and generated by Jabbah.
Starting from this XML plan representation, we have implemented as an extension of
Jabbah a translation process that automatically generates XPDL processes which can be
directly executed in a BPM runtime engine and users can interact with them on an un-
derlying BPM console (see [8] for more details). This process has three main steps: (1)
generation of XPDL DataFields and Participants from the problem and domain files;
(2) generation of XPDL activities from the information about actions, temporal con-
straints and metadata in the plan; (3) generation of XPDL transitions from the order
dependencies between the actions of the plan.
3 Conclusions
Jabbah fulfills, by using AI P&S knowledge engineering techniques, the needs of knowl-
edge workers not yet completely covered by BPM technologies, in the management
of dynamic, adaptable processes. The main innovative aspects of this framework are
both, the fully automated transformation from a business process model (represented in
XPDL) into an HTN planning domain and the translation from plan representation into
an executable process format, what allows to directly execute the result of a planning
process in standard BPM runtime engines. The framework described presents signifi-
cant advances in the field of BPM, since the seamless integration of the above explained
techniques leverages the BPM life-cycle, allowing it to automatically carry out adaptive
case generation (starting from an initial process model represented in BPM standard
languages), and to execute cases using standard BPM technologies.
To date, JABBAH has been tested in two case studies drawn from different do-
mains, e-learning and e-health. The first model represents the whole process to develop
and deploy a specific course within the e-learning center at the University of Granada.
So, having an incoming course request, as well as some available workers with different
capabilities each, we want to assign an activity to every worker, so that a plan over time
can be obtained, providing the e-learning managers information that helps to do antic-
ipated decision-making upon the course request. The second one represents a general
care-process starting from a patient admitted into the hospital and finishing when the
health insurance billing for this patient takes place. In this second process, we can better
observe how process planning is carried out, given different input parameters which can
vary in real situations (e.g. is it an emergency? does it need an urgent operation?) and
how our tool is able to generate different process instances according to them.
Source code and details about the Jabbah framework and the case studies com-
mented are available at its website1 . A demonstration screencast about its operation can
be watched in the ”Screenshots” subsection of the website, or directly in Youtube2 .
References
1. Bae, J., Bae, H., Kang, S., Y.Kim: ”Automatic Control of Workflow Processes Using ECA
Rules”. IEEE Transactions on Knowlegde and Data Engineering 16(8) (2004)
2. Bresina, J.L., Jonsson, A.K., Morris, P., Rajan, K.: Activity planning for the mars exploration
rovers. In: Proceedings of the ICAPS05. pp. 40–49 (2005)
3. Castillo, L., Fdez-Olivares, J., Garcı́a-Pérez, O., Garzón, T., Palao, F.: Reducing the impact
of ai planning on end users. In: ICAPS 2007, Workshop on Moving Planning and Scheduling
Systems into the Real World. pp. 40–49 (2007)
4. Castillo, L., Fdez-Olivares, J., Garcı́a-Pérez, O., Palao, F.: Efficiently handling temporal
knowledge in an HTN planner. In: Proceeding of ICAPS06. pp. 63–72 (2006)
5. Dayal, U., Hsu, M., Ladin, R.: Business process coordination: State of the art, trends, and
open issues. In: Proceedings of the 27th VLDB Conference (2001)
6. Fdez-Olivares, J., Castillo, L., Garcı́a-Pérez, O., Palao, F.: Bringing users and planning tech-
nology together. Experiences in SIADEX. In: Proceedings ICAPS06. pp. 11–20 (2006)
7. Fdez-Olivares, J., Castillo, L., Cozar, J., Garcia-Perez, O.: Supporting clinical processes and
decisions by hierarchical planning and scheduling. Computational Intelligence To Appear
(2010)
8. Fdez-Olivares, J., González-Ferrer, A., Sanchez-Garzón, I., Castillo, L.: Integrating plans
into BPM technologies for human-centric process execution. In: ICAPS 2010. Proceedings
of Workshop on Knowledge Engineering for Planning and Scheculing (KEPS) (2010)
9. Fox, M., Long, D.: PDDL2-1: an extension to PDDL for expressing temporal planning do-
mains. Tech. rep., University of Durham, UK (2001)
10. Ghallab, M., Nau, D., Traverso, P.: Automated Planning. Theory and Practice. Morgan Kauf-
mann (2006)
11. González-Ferrer, A., Fdez-Olivares, J., Castillo, L.: ”JABBAH: A Java Aplication Frame-
work for the Translation between Business Process Models and HTN”. In: Proceedings of
ICKEPS Competition (2009)
12. Leoni, M.D.: Adaptive process management in highly dynamic and pervasive scenarios. In:
YR-SOC 2009. pp. 83–97
13. Nau, D., Au, T., Ilghami, O., Kuter, U., Murdock, J.W., Wu, D., Yaman, F.: SHOP2: An HTN
Planning System. Journal of Artificial Intelligence Research 20, 379–404 (2003)
14. Sacerdoti, E.D.: The nonlinear nature of plans. In: Proceedings of IJCAI 1975. pp. 206–214
(1975)
15. WfMC: Group on adaptive case management (2010), http://www.xpdl.org/nugen/
p/adaptive-case-management/public.htm
1
http://sites.google.com/site/bpm2hth
2
http://www.youtube.com/watch?v=FOHYsMWvS1c