Defining system requirements in engineering design is a complex and critical task. Among various sources of information, requirements must be accurately identified to ensure successful outcomes. Large Language Models (LLMs) offer designers access to extensive knowledge and can generate conceptual design solutions in natural language, providing valuable support in early-stage design. This study explores the potential of LLMs in developing system requirements through an automotive case study focussed on valve design. A mixed-methods approach combined quantitative surveys and qualitative interviews with industry professionals who manage requirements. The dataset included human-created requirements, which were compared to AI-generated requirements and evaluated using four quality criteria: specificity, functionality, target values, and verifiability. Interviews further revealed current workflow challenges, including ambiguity, requirement overloading, difficulty defining early-stage requirements, and frequent changes. While participants acknowledged these challenges, they also recognised the benefits and limitations of AI support. Results suggest that AI-generated requirements can make the process more manageable, act as a collaborative partner, and improve efficiency. The quantitative findings are exploratory and intended to reveal indicative patterns rather than support generalisable statistical conclusions. Although AI may overlook certain details, findings highlight significant potential for its refinement to enhance requirement accuracy. This work contributes to understanding how AI can support one of the most demanding aspects of engineering design.
Rahmanpour et al. (Wed,) studied this question.