Generative AI with Modeling and Simulation of Activity and Flow-Based Diagrams

In systems engineering and model-based design, the complexity and interrelationships across different system elements always demand continuous elaboration and expansion in various overlapping domains. We examine how such a phenomenon can be assisted with generative AI and benefit from large language models (LLMs), such as GPT-4. We demonstrate ways of incorporating generated text and outputs from LLMs into the modeling process. The approach can customize the GPT-4 model with an activity metamodel specified in Eclipse Ecore or predefined activity diagrams encoded in a textual format for learning from instances. Alternatively, the descriptive text from the LLM can be provided as input to a parser, resulting in an activity that can be readily transformed into a discrete event system specification (DEVS) model with simulation capability. We will discuss how the process can be enhanced in a simulation environment, thus offering the opportunity to examine a variety of scenarios and arguments for incorporating generative AI or general AI as a collaborative agent in the domain of interest. One scenario could begin with a simplified text describing a generic process, yielding an approximate representation as a starting point for further elaboration by modelers to a complex specification through a systematic, guided, and well-defined framework. We demonstrate the approach with activity and flow-based diagrams in a manner applicable to SysML, UML, and systems engineering at large.