An Enhanced Design Concept Evaluation Approach Incorporating Large Language Models and Heterogeneous Preference Fusion

Abstract Design concept evaluation (DCE) is a crucial stage in new product development, significantly influencing downstream design and manufacturing processes. However, the subjectivity, heterogeneity, and potential conflicts in designers' preference expressions present substantial challenges to reliable evaluation. To address these issues, an intelligent DCE approach that integrates large language models (LLMs) and heterogeneous designer preferences is proposed. Initially, the patent application numbers and International Patent Classification (IPC) are extracted from the patent database, and a preliminary screening approach is established using a few-shot learning-based LLM to identify high-quality design concepts. Second, the remaining design concepts are evaluated based on designers' heterogeneous preferences. A preference fusion model and a two-layer consensus-reaching algorithm are developed to aggregate diverse preferences and mitigate conflicts. Third, the optimal design concept is selected based on the preference ranking organization method for enrichment evaluations Ⅱ (PROMETHEE Ⅱ) and 0–1 programming, which considers the compatibility among subfunctions. A case study involving an automatic bar peeling machine is conducted to demonstrate the feasibility of the proposed approach. The robustness analysis and approach comparison confirm that the proposed approach can effectively enhance the objectivity and credibility of DCE.