A case-based knowledge graph with reinforcement learning for intelligent design approach of complex product

Abstract

Amidst the challenge of striving for rapid production and cost reduction, enterprises exhibit a growing demand for inventive, efficient, and high-caliber solutions in product design. The design field has followed the integration of Knowledge Graph and Artificial Intelligence technologies, and attempts have been made to address the demands of rapidity and intelligence. However, effectively leveraging knowledge to achieve rapid product generation remains a challenge. Thus, this paper proposes a case-based knowledge graph with reinforcement learning for the intelligent design approach of complex products, intending to tackle the difficulty of efficiently representing case knowledge of complex products and identifying key modification points in case reconstruction. Knowledge graph technology integrates and defines the features and component data from historical cases intuitively and structuredly. Similarity calculation is employed to facilitate case retrieval and the acquisition of configurable components. Combining the component relationships of the case graphs provides reward values for further adopted reinforcement learning, thereby aiding designers in identifying critical components to reconstruct the case further and achieving a balance between configuration efficiency and product quality. Finally, radar design is utilised as an exemplar to validate the efficacy of the proposed method.