GenAI for Simulation Model in Model-Based Systems Engineering

TL;DR

This work presents a GenAI-driven framework for MBSE that automatically generates simulation models of system physical properties from product design documents by aligning DEVS-based modeling with an X language and scalable templates. It combines NER-BERT-based information extraction to identify system composition, Transformer-based code generation (with LoRA fine-tuning on CodeQwen) for atomic-class behavior, and modular template-driven assembly to produce consistent, long-context code. A tailored evaluation scheme using Degree of Error and Model Consistency, weighted by entropy, provides quantitative quality assessments beyond traditional code-accuracy metrics. The aircraft electrical system case demonstrates improved model quality and integration of GenAI into MBSE workflows, while highlighting limitations related to prior knowledge bias and the need for richer datasets and lifecycle coverage.

Abstract

Generative AI (GenAI) has demonstrated remarkable capabilities in code generation, and its integration into complex product modeling and simulation code generation can significantly enhance the efficiency of the system design phase in Model-Based Systems Engineering (MBSE). In this study, we introduce a generative system design methodology framework for MBSE, offering a practical approach for the intelligent generation of simulation models for system physical properties. First, we employ inference techniques, generative models, and integrated modeling and simulation languages to construct simulation models for system physical properties based on product design documents. Subsequently, we fine-tune the language model used for simulation model generation on an existing library of simulation models and additional datasets generated through generative modeling. Finally, we introduce evaluation metrics for the generated simulation models for system physical properties. Our proposed approach to simulation model generation presents the innovative concept of scalable templates for simulation models. Using these templates, GenAI generates simulation models for system physical properties through code completion. The experimental results demonstrate that, for mainstream open-source Transformer-based models, the quality of the simulation model is significantly improved using the simulation model generation method proposed in this paper.

Figures (9)

• Figure 1: Based on XDEVS (an extension of DEVS), X language defines various classes, including discrete, continuous, and agent classes, and supports both graphical and textual representations. X language supports modeling continuous, discrete, and hybrid models and enables the compilation and simulation of models using X language development tools, XLab.
• Figure 2: Elements in X language classes can be mapped to corresponding elements in DEVS bib56. Taking the coupled model of DEVS as an example, the functions of $EIC$ and $EOC$ are implemented through the keywords Port and Connection in the couple class of X language. The function of $IC$ is implemented using the keyword Connection alone.
• Figure 3: The overall technical implementation framework is divided into three parts: Simulation Model Generation Method Based on Scalable Templates and Transformer-based Models, Training of Language Models for Simulation Model Generation, Evaluation Metrics for Simulation Models Generated by Transformer-based Models.
• Figure 4: The NER-BERT architecture comprises BERT, a fully connected layer, and a softmax layer. Multiple product design documents and random noise data are transformed into a training set through WordPiece and manual tagging. This dataset was subsequently used to fine-tune NER-BERT.
• Figure 5: This is an example of generating an X language simulation model. We extracted several X language models from X language model library and used them as samples to construct prompts using Few-shot learning. ChatGPT generates new X language models based on these prompts.