Using Large Language Models for Fluid Power System Design

抄録

The advent of Large Language Models (LLMs) has demonstrated considerable promise across various domains, including engineering system design, owing to their capabilities as few-shot or even single-shot learners. This paper investigates the integration of LLMs in the architectural design of fluid power systems for construction machinery. The primary contribution is the development of a methodology that transforms natural language inputs into formal architectural definitions, utilising micro-templates within prompts to enhance the repeatability and consistency of outputs.

As an additional step, LLMs can be instructed to generate Python code that produces system architectures based on user-defined inputs, thereby operating within a more constrained but reliable design space. The iterative refinement process facilitated by LLMs enables the progressive expansion and optimisation of design spaces, offering significant potential to automate and accelerate the early phases of engineering design.