Modeling and simulation have been established as indispensable tool in science and engineering, offering means to explore and analyze complex systems, thereby enhancing the prediction and understanding of real-world phenomena. Developing valid models that accurately reproduce these phenomena is an intricate process, involving steps of model building intertwined with various simulation experiments for model calibration, validation, and analysis.
Especially conducting simulation experiments is challenging as their specification and execution vary depending, for example, on the diverse experiment types, methods, and tools required. This dissertation addresses the central question of how to support modelers in this intricate process, specifically, by automatically generating and executing simulation experiments. A pivotal strategy identified for achieving this goal is the generation of simulation experiments through reuse and adaption. The strategy can be subdivided into three parts, and the dissertation is organized accordingly.
The first part contributes to the explicit and automated specification of simulation experiments, employing model-driven engineering. In this approach, metamodels are crucial for formalizing the central ingredients of a simulation experiment and for defining a domain-specific language that allows representing simulation experiments in a tool-agnostic manner. This enables the automatic generation, transformation, and adaption of executable experiment specifications via a model-driven engineering pipeline.
The second part underlines the importance of context for understanding and conducting simulation studies, and particularly simulation experiments. A comprehensive and formalized definition of the conceptual model is introduced, which subsumes typical early-stage artifacts of a simulation study, from research objectives and assumptions to information sources, enriched with metadata. The conceptual model can be complemented by provenance, which is concerned with the process of generating and using the various artifacts. Formal approaches for representing provenance information are discussed that can illuminate the entire story of a simulation study, including the interrelations between artifacts, activities of model building and simulation experiments, and related studies. Open model databases are investigated as a key means for distributing, editing, and interconnecting provenance graphs and the conceptual model, ensuring their computational accessibility.
The third part focuses on utilizing the formal, machine-accessible representations to facilitate the automatic interpretation and exploitation of context information. This context information is integrated with the model-driven approach for experiment specification and execution to form the key components of the Reuse and Adapt framework for Simulation Experiments (RASE). In this framework, provenance patterns and inference rules enable the automatic initiation of an experiment generator depending on the last activities of a modeler, automatic identification of earlier simulation experiments to reuse, and their adaption to the context of a new simulation model. This supports the iterative nature of the modeling and simulation life cycle, enabling scenarios such as automatic regression testing and cross-validation of related models.
The results showcase how modelers can benefit from the developed approaches, emphasizing systematic, effective, and methodologically rigorous simulation studies while preserving user flexibility and control. Demonstrations in open-source software prototypes and diverse case studies, spanning stochastic discrete-event simulations, virtual prototyping, finite element analysis, and agent-based simulations, highlight the versatility and applicability of the proposed framework.
Finally, the dissertation identifies new research avenues towards fully automating simulation experiments as well as the entire modeling and simulation life cycle.