Written by Pingfei Jiang, Ji Han and Saeema Ahmed-Kristensen
Design is fundamental to product development and is widely recognised as a key driver for supporting the transition to a circular economy.
As a problem-solving process, design relies heavily on knowledge retrieval and generation. Large language models (LLMs), such as ChatGPT, have demonstrated substantial capabilities in these areas, suggesting their potential to support sustainable product design. However, LLMs alone often produce ambiguous and overly general responses, limiting their effectiveness in delivering valuable sustainability insights. To address this limitation, retrieval-augmented generation (RAG), which integrates domain-specific information retrieval with generative modelling, has emerged as a promising approach to improve output accuracy and contextual relevance.
This study introduces a novel two-stage RAG framework specifically designed to enhance sustainable product development by incorporating two external knowledge bases: one dedicated to systems design principles and the other to sustainable design strategies. This framework equips designers with practical sustainable design guidelines that align naturally with the design thinking process, requiring minimal prior sustainability expertise, an improvement over current approaches that depend heavily on specialised knowledge or assessment tools typically applied in later development stages.
Expert-evaluated case studies demonstrate that this RAG-based framework covers 2.7 times more relevant product design specification topics than non-RAG approaches, with noticeable improvements across all performance metrics, particularly in product Desirability and Use lifecycle stage considerations. This research has implications for accelerating circular economy transitions by embedding sustainability considerations at the conceptual design stage, potentially transforming how designers approach sustainable product innovation with relatively low implementation costs.
