Multi-perspective trace alignment in process mining with automated planning

This thesis advances planning-based techniques for multi-perspective process analysis and execution support in Business Process Management. Starting from the observation that trace alignment, one of the most powerful conformance checking techniques, can be formulated as a cost-optimal planning problem in AI, the thesis systematically extends this formulation beyond pure control-flow deviations. In particular, context-dependent cost models are introduced to capture deviations whose severity varies according to the execution context, and timed trace alignment is formalized by representing temporal constraints through 1-clock Deterministic Timed Automata and reducing the alignment problem to numeric planning. These techniques are further extended towards runtime execution support through the notion of framed autonomy, which enables AI-augmented Business Process Management Systems to autonomously progress partially executed processes while respecting a set of procedural and declarative constraints at minimum violation cost. The practical applicability of the proposed approach is demonstrated through an application to Cybersecurity Incident Management compliance assessment. Finally, to address the bottleneck of formal domain modeling, an LLM-based pipeline for generating PDDL domains from natural language descriptions is presented. Together, these contributions establish a unified automata-based and planning-based framework for multi-perspective conformance checking, prescriptive process reasoning, and accessible domain modeling.