Mining sound free-choice workflow nets: beyond block structures
Year: 2026
Type: dissertation
Abstract: Organizations increasingly rely on information systems to execute and manage their business processes, resulting in the continuous generation of large volumes of event data. Process mining exploits this data to reconstruct, analyze, and improve business processes, thereby addressing the gap between prescribed process designs and their actual execution. Among the core tasks of process mining, process discovery plays a central role, as the quality of discovered models directly determines the effectiveness of subsequent conformance checking and process enhancement. However, many existing discovery techniques impose block-structured constraints, which limit their ability to represent complex, intertwined control-flow behavior observed in real-world processes. This thesis investigates the discovery and analysis of process models from event logs with a focus on \emph{sound free-choice workflow nets beyond block structures}.Sound free-choice workflow nets provide a balance between expressive power and formal analyzability, offering strong behavioral guarantees such as soundness while supporting rigorous analysis. Building on a characterization of this model class, the thesis develops synthesis rules-based discovery techniques that construct sound free-choice workflow nets directly from event logs without enforcing block-structured representations. The proposed methods ensure soundness and the free-choice property by design, enabling the discovery of expressive, non-hierarchical models that better reflect observed process behavior. To support the analysis of complex, non-block-structured models, this thesis also introduces a decomposition technique tailored to sound free-choice workflow nets. The approach partitions large nets into smaller fragments while strictly preserving soundness and the free-choice property, thereby enabling modular analysis and scalable application of diagnostic techniques. The proposed methods are grounded in formal definitions and correctness arguments and are supported by tool implementations. The empirical evaluation on benchmark and synthetic event logs demonstrates that the proposed techniques achieve a favorable balance between fitness and simplicity when compared to block-structured discovery approaches, particularly for processes exhibiting complicated control-flow patterns. Overall, this thesis provides a formally grounded methodology for discovering and analyzing expressive workflow nets that go beyond block-structured representations. By combining synthesis-based discovery, decomposition techniques, and formal guarantees, it contributes toward more reliable and theoretically principled process discovery methods while outlining directions for addressing remaining challenges related to scalability, noise, and expressiveness.
Source: RWTH Publications (RWTH Aachen)
Author: Huang, Tsung-Hao
Institution: RWTH Aachen University
Language: English
Subfield: Management Information Systems
Domain: Social Sciences
Open Access status: green
OpenAlex ID: W7222516359
Cites:
Cited by:
Related to: