Design deduction for multi-dimensional evaluation: a multi-agent collaboration based framework

Abstract

Design deduction and design reasoning are core topics in the field of design automation, aiming to simulate the design thinking process through computational models and assist in the generation and optimization of solutions. Most of the existing research focuses on single-objective optimization or rule-driven design suggestions, lacking a multi-dimensional systematic evaluation of design works. This leads to insufficient comprehensive analysis of the reasoning results in terms of rationality, feasibility and risk, which limits their application in complex innovative design. At present, the work of automatic design deduction is often limited to a single evaluation dimension and lacks the ability of multi-objective collaborative deduction. Moreover, most systems rely on static rules and are difficult to adapt to a dynamic and open design context. Meanwhile, the existing methods have obvious shortcomings in cross-domain knowledge fusion and forward-looking risk prediction, resulting in limited practicality and innovation of the deduction suggestions. This paper proposes a design deduction framework based on multi-agent systems (MAS), which includes three Agent modules for evaluation: (1) Rationality evaluation Agent: Based on design theory and domain knowledge, it assesses the consistency of design logic, user experience and contextual adaptability; (2) Feasibility assessment Agent: By integrating engineering constraints and technical parameters, analyze the feasibility of manufacturing processes, costs, and resources; (3) Risk assessment Agent: Through historical data and simulation prediction, identify potential risks in technology, market and ethics. Each Agent debates and negotiates through a competition-collaboration mechanism. The central coordinator comprehensively outputs multi-dimensional optimization strategies to achieve dynamic iterative design deduction. Experiments show that, compared with the single-dimensional evaluation system, this framework has achieved relevant improvements in dimensions such as the adoption rate of optimization suggestions and the accuracy rate of risk early warning in the derivation of concept products and architectural design schemes. The collaborative deduction mechanism effectively shortens the design iteration cycle and demonstrates stronger strategy generation capabilities and contextual adaptability in cross-domain innovative design.

Keywords: Design Deduction, Multi-dimensional Evaluation, Multi-agent Systems

DOI: 10.54941/ahfe1007529