AI Planning, Machine Learning and Process Mining have so far developed into separate research fields. At the same time, many interesting concepts and insights have been gained at the intersection of these areas in recent years. For example, the behavior of future processes is now comprehensively predicted with the aid of Machine Learning. For the practical application of these findings, however, it is also necessary not only to know the expected course, but also to give recommendations and hints for the achievement of goals, i.e. to carry out comprehensive process planning. At the same time, an adequate integration of the aforementioned research fields is still lacking. In this article, we present a research project in which researchers from the AI and BPM field work jointly together. Therefore, we discuss the overall research problem, the relevant fields of research and our overall research framework to automatically derive process models from executional process data, derive subsequent planning problems and conduct automated planning in order to adaptively plan and execute business processes using real-time forecasts.
Business Process Management (BPM) plays a vital role for organizations. The consistent design,
implementation, execution and monitoring of business processes promises eficiency increases
and cost reduction. In today’s organizations, BPM follows the paradigm of continuous process
improvement. After a process is designed, implemented and executed, the process execution
data is retrospectively analyzed with methods such as Process Mining in order to optimize
process design in a cyclic manner. In this framework however, process agility (e.g. process
variants) has to be defined at design time in order to achieve full process automation, which is
typically not feasible in real-world scenarios. Approaches using Process Mining and Machine
Learning (ML) to overcome this problem exist, however stakeholders are left to their own
on how to use these insights [
Our work is positioned within this research gap at the intersection of AI Planning, ML and Process Mining, as depicted in Figure 1. We propose developing new methods to achieve online automated planning leading to an agile execution of business processes. In that regard, we aim at answering the following research question: How is an automated planning of processes achieved using automatically derived process models and automatically constructed planning models?
We propose a procedure model divided into five phases leading from executional process data to
executable planning models, as depicted in the right part of Figure 1. The general starting point
is an ERP system providing process execution data in shape of relational data and execution
logs. The second stage deals with extracting a process execution log covering the occurring
cases and events. Based on that, the next important step is to automatically derive a process
model. The key aspect hereby is that the resulting model entails enough process details to allow
for AI planning, e.g. not only the general process flow, but also implicit domain knowledge,
involved resources or potential resource allocation constraints. Such elements can then be
used to automatically construct a sophisticated planning domain in subsequent phases. In this
regard we also propose taking a broader view and starting to think about mining entire system
models which describe the architecture of (hierarchical) systems including their static as well as
dynamic elements based on distributed runs observable from event logs [
We designed diferent application cases to develop and evaluate our approach, most prominently
a classic flow-shop scheduling problem shown in Figure 2 modeled using Heraklit [
In this scenario, a set of diferent products have to be produced on a production line consisting of diferent machine stations and workers to operate them, where each product has to cross all stations. Figure 2 shows an exemplary setup consisting of 10 stations and 10 workers. The scenario includes several variable factors, e.g. the process execution time of one production step depends on the underlying product, station as well as the operating worker with a distinct experience level. The general steps to perform at each station can difer, adding another dimension to the planning problem. On top of that, the process includes walking times from one station to another whenever necessary, which in turn is dependent on the layout of the production line, i.e. how the individual stations are distributed.
Regarding phases two and three, multiple modeling frameworks can be considered. While XES
is a well-known standard for event logs, we came to the conclusion that such representation
might not sufice due to not providing enough information about involved resources necessary
for planning domains. Heraklit on the other hand is a modeling language based on symbolic
Petri Nets which we regard as a good foundation for subsequent planning procedures. It also
allows us to model entire system models as described in section 2 including dependencies
between individual subsystems and agents. For more details about Heraklit, we refer to [
Several advances have been made regarding the latter parts of the framework, e.g.
“domainspecific Constraint Patterns” in the context of Constraint Programming [
We also propose a novel GPU-assisted approach for on-the-fly generation of training samples
for Deep-Q learning [
Based on this approach, we developed Deep-Q-Agents methods for optimization. To achieve convincing results using such techniques, it is necessary to set up domain specific configurations, which in turn requires domain knowledge. This however also requires a human assessment of these configurations. To overcome this problem, we suggest an automated exploration of the underlying search space as a heuristic to find potentially optimal configurations.
Several research initiatives in the fields of AI Planning and Process Mining in the context of BPM have been made so far. However, there are no satisfying results when it comes to planning and agile executions of business processes in an automated fashion. To this end, we propose an iterative procedure model build upon executional process data to automatically derive process models and automatically infer solvable planning problems to plan and optimize business processes. Up until now, several concepts have been implemented and advancements addressing individual parts of the procedure model have been made. The next main step is to adequately combine these individual concepts and put the proposed "round-trip" into practice.
This work is part of the research project APPaM (Grant 01IW20006), which is partly funded by the Federal Ministry of Education and Research (BMBF).