Generation of MATLAB Simscape Models Using Large Language Models for Training Reliability Evaluation Methods
Patrick Hummel, Joachim Grimstad, Yuchen Xia, Andrey S. Morozov · 2024
Abstract The approach presented in this paper makes use of the natural language processing capabilities of large language models in order to generate MATLAB Simscape models based on specification texts. This process is based on a well-defined state machine and begins with the creation of abstract system models that may be iterated on using natural language. The abstract models are then converted to more detailed models by adding parameters and blocks required for simulation. Fault tolerance may optionally be improved by automatically applying design patterns that increase redundancy using voter and comparator blocks, among others. The abstract models are then converted into MATLAB Simscape models using traditional algorithms. The generated models can be used to train artificial intelligence systems designed for reliability evaluation. A selection of publicly available large language models is systematically evaluated based on capability regarding this task. A software prototype demonstrates the feasibility of using generative artificial intelligence to assist with the modeling of systems in MATLAB Simscape as a proof-of-concept. Current limitations and future research potential is discussed.